diff options
Diffstat (limited to 'prism_compile.c')
| -rw-r--r-- | prism_compile.c | 7840 |
1 files changed, 4506 insertions, 3334 deletions
diff --git a/prism_compile.c b/prism_compile.c index 14cf016f16..45e1de8a9c 100644 --- a/prism_compile.c +++ b/prism_compile.c @@ -1,9 +1,12 @@ #include "prism.h" +#include "ruby/version.h" + +#include <fcntl.h> /** * This compiler defines its own concept of the location of a node. We do this * because we want to pair line information with node identifier so that we can - * have reproducable parses. + * have reproducible parses. */ typedef struct { /** This is the line number of a node. */ @@ -101,6 +104,7 @@ pm_iseq_add_setlocal(rb_iseq_t *iseq, LINK_ANCHOR *const seq, int line, int node else { ADD_ELEM(seq, (LINK_ELEMENT *) new_insn_body(iseq, line, node_id, BIN(setlocal), 2, INT2FIX((idx) + VM_ENV_DATA_SIZE - 1), INT2FIX(level))); } + update_lvar_state(iseq, level, idx); if (level > 0) access_outer_variables(iseq, level, iseq_lvar_id(iseq, idx, level), Qtrue); } @@ -138,39 +142,127 @@ pm_iseq_add_setlocal(rb_iseq_t *iseq, LINK_ANCHOR *const seq, int line, int node #define PM_COMPILE_NOT_POPPED(node) \ pm_compile_node(iseq, (node), ret, false, scope_node) -#define PM_SPECIAL_CONSTANT_FLAG ((pm_constant_id_t)(1 << 31)) -#define PM_CONSTANT_AND ((pm_constant_id_t)(idAnd | PM_SPECIAL_CONSTANT_FLAG)) -#define PM_CONSTANT_DOT3 ((pm_constant_id_t)(idDot3 | PM_SPECIAL_CONSTANT_FLAG)) -#define PM_CONSTANT_MULT ((pm_constant_id_t)(idMULT | PM_SPECIAL_CONSTANT_FLAG)) -#define PM_CONSTANT_POW ((pm_constant_id_t)(idPow | PM_SPECIAL_CONSTANT_FLAG)) +// Direct-indexed lookup table. -1 means "not present". +#define PM_INDEX_LOOKUP_TABLE_INIT { .values = NULL, .capacity = 0, .owned = false } -#define PM_NODE_START_LOCATION(parser, node) \ - ((pm_node_location_t) { .line = pm_newline_list_line(&(parser)->newline_list, ((const pm_node_t *) (node))->location.start, (parser)->start_line), .node_id = ((const pm_node_t *) (node))->node_id }) +static inline void +pm_index_lookup_table_init(pm_index_lookup_table_t *table, int constants_size, rb_iseq_t *iseq) +{ + int capacity = constants_size + PM_INDEX_LOOKUP_SPECIALS; + table->values = compile_data_alloc2_type(iseq, int, capacity); + memset(table->values, -1, capacity * sizeof(int)); + table->capacity = capacity; + table->owned = false; +} -#define PM_NODE_END_LOCATION(parser, node) \ - ((pm_node_location_t) { .line = pm_newline_list_line(&(parser)->newline_list, ((const pm_node_t *) (node))->location.end, (parser)->start_line), .node_id = ((const pm_node_t *) (node))->node_id }) +/** + * Cached line lookup that avoids repeated binary searches. Since the compiler + * walks the AST roughly in source order, consecutive lookups tend to be for + * nearby byte offsets. We cache the last result index in the scope node and + * try a short linear probe from there before falling back to binary search. + */ +static inline pm_line_column_t +pm_line_offset_list_line_column_cached(const pm_line_offset_list_t *list, uint32_t cursor, int32_t start_line, size_t *last_line) +{ + size_t hint = *last_line; + size_t size = list->size; + const uint32_t *offsets = list->offsets; -#define PM_LOCATION_START_LOCATION(parser, location, id) \ - ((pm_node_location_t) { .line = pm_newline_list_line(&(parser)->newline_list, (location)->start, (parser)->start_line), .node_id = id }) + RUBY_ASSERT(hint < size); -#define PM_NODE_START_LINE_COLUMN(parser, node) \ - pm_newline_list_line_column(&(parser)->newline_list, ((const pm_node_t *) (node))->location.start, (parser)->start_line) + /* Check if the cursor is on the same line as the hint. */ + if (offsets[hint] <= cursor) { + if (hint + 1 >= size || offsets[hint + 1] > cursor) { + *last_line = hint; + return ((pm_line_column_t) { + .line = ((int32_t) hint) + start_line, + .column = cursor - offsets[hint] + }); + } -#define PM_NODE_END_LINE_COLUMN(parser, node) \ - pm_newline_list_line_column(&(parser)->newline_list, ((const pm_node_t *) (node))->location.end, (parser)->start_line) + /* Linear scan forward (up to 8 lines before giving up). */ + size_t limit = hint + 9; + if (limit > size) limit = size; + for (size_t idx = hint + 1; idx < limit; idx++) { + if (offsets[idx] > cursor) { + *last_line = idx - 1; + return ((pm_line_column_t) { + .line = ((int32_t) (idx - 1)) + start_line, + .column = cursor - offsets[idx - 1] + }); + } + if (offsets[idx] == cursor) { + *last_line = idx; + return ((pm_line_column_t) { ((int32_t) idx) + start_line, 0 }); + } + } + } + else { + /* Linear scan backward (up to 8 lines before giving up). */ + size_t limit = hint > 8 ? hint - 8 : 0; + for (size_t idx = hint; idx > limit; idx--) { + if (offsets[idx - 1] <= cursor) { + *last_line = idx - 1; + return ((pm_line_column_t) { + .line = ((int32_t) (idx - 1)) + start_line, + .column = cursor - offsets[idx - 1] + }); + } + } + } -#define PM_LOCATION_START_LINE_COLUMN(parser, location) \ - pm_newline_list_line_column(&(parser)->newline_list, (location)->start, (parser)->start_line) + /* Fall back to binary search. */ + pm_line_column_t result = pm_line_offset_list_line_column(list, cursor, start_line); + *last_line = (size_t) (result.line - start_line); + return result; +} -static int -pm_node_line_number(const pm_parser_t *parser, const pm_node_t *node) +/** + * The same as pm_line_offset_list_line_column_cached, but returning only the + * line number. + */ +static inline int32_t +pm_line_offset_list_line_cached(const pm_line_offset_list_t *list, uint32_t cursor, int32_t start_line, size_t *last_line) { - return (int) pm_newline_list_line(&parser->newline_list, node->location.start, parser->start_line); + return pm_line_offset_list_line_column_cached(list, cursor, start_line, last_line).line; } +#define PM_NODE_START_LOCATION(node) \ + ((pm_node_location_t) { .line = pm_line_offset_list_line_cached(scope_node->line_offsets, ((const pm_node_t *) (node))->location.start, scope_node->start_line, &scope_node->last_line), .node_id = ((const pm_node_t *) (node))->node_id }) + +#define PM_NODE_END_LOCATION(node) \ + ((pm_node_location_t) { .line = pm_line_offset_list_line_cached(scope_node->line_offsets, ((const pm_node_t *) (node))->location.start + ((const pm_node_t *) (node))->location.length, scope_node->start_line, &scope_node->last_line), .node_id = ((const pm_node_t *) (node))->node_id }) + +#define PM_LOCATION_START_LOCATION(location, id) \ + ((pm_node_location_t) { .line = pm_line_offset_list_line_cached(scope_node->line_offsets, (location)->start, scope_node->start_line, &scope_node->last_line), .node_id = id }) + +#define PM_NODE_START_LINE_COLUMN(node) \ + pm_line_offset_list_line_column_cached(scope_node->line_offsets, ((const pm_node_t *) (node))->location.start, scope_node->start_line, &scope_node->last_line) + +#define PM_NODE_END_LINE_COLUMN(node) \ + pm_line_offset_list_line_column_cached(scope_node->line_offsets, ((const pm_node_t *) (node))->location.start + ((const pm_node_t *) (node))->location.length, scope_node->start_line, &scope_node->last_line) + +#define PM_LOCATION_START_LINE_COLUMN(location) \ + pm_line_offset_list_line_column_cached(scope_node->line_offsets, (location)->start, scope_node->start_line, &scope_node->last_line) + static int pm_location_line_number(const pm_parser_t *parser, const pm_location_t *location) { - return (int) pm_newline_list_line(&parser->newline_list, location->start, parser->start_line); + return (int) pm_line_offset_list_line_column(pm_parser_line_offsets(parser), location->start, pm_parser_start_line(parser)).line; +} + +/** + * Cached variants that use the scope node's hint for fast lookups during + * compilation (where access patterns are roughly sequential). + */ +static inline int +pm_node_line_number_cached(const pm_node_t *node, pm_scope_node_t *scope_node) +{ + return (int) pm_line_offset_list_line_cached(scope_node->line_offsets, node->location.start, scope_node->start_line, &scope_node->last_line); +} + +static inline int +pm_location_line_number_cached(const pm_location_t *location, pm_scope_node_t *scope_node) { + return (int) pm_line_offset_list_line_cached(scope_node->line_offsets, location->start, scope_node->start_line, &scope_node->last_line); } /** @@ -185,24 +277,25 @@ parse_integer_value(const pm_integer_t *integer) result = UINT2NUM(integer->value); } else { - VALUE string = rb_str_new(NULL, integer->length * 8); - unsigned char *bytes = (unsigned char *) RSTRING_PTR(string); - - size_t offset = integer->length * 8; - for (size_t value_index = 0; value_index < integer->length; value_index++) { - uint32_t value = integer->values[value_index]; - - for (int index = 0; index < 8; index++) { - int byte = (value >> (4 * index)) & 0xf; - bytes[--offset] = byte < 10 ? byte + '0' : byte - 10 + 'a'; - } - } - - result = rb_funcall(string, rb_intern("to_i"), 1, UINT2NUM(16)); + // The pm_integer_t stores values as an array of uint32_t in + // least-significant-word-first order (base 2^32). We can convert + // directly to a Ruby Integer using rb_integer_unpack, avoiding the + // overhead of constructing a hex string and calling rb_funcall. + result = rb_integer_unpack( + integer->values, + integer->length, + sizeof(uint32_t), + 0, + INTEGER_PACK_LSWORD_FIRST | INTEGER_PACK_NATIVE_BYTE_ORDER + ); } if (integer->negative) { - result = rb_funcall(result, rb_intern("-@"), 0); + result = rb_int_uminus(result); + } + + if (!SPECIAL_CONST_P(result)) { + RB_OBJ_SET_SHAREABLE(result); // bignum } return result; @@ -223,7 +316,11 @@ parse_integer(const pm_integer_node_t *node) static VALUE parse_float(const pm_float_node_t *node) { - return DBL2NUM(node->value); + VALUE val = DBL2NUM(node->value); + if (!FLONUM_P(val)) { + RB_OBJ_SET_SHAREABLE(val); + } + return val; } /** @@ -237,7 +334,8 @@ parse_rational(const pm_rational_node_t *node) { VALUE numerator = parse_integer_value(&node->numerator); VALUE denominator = parse_integer_value(&node->denominator); - return rb_rational_new(numerator, denominator); + + return rb_ractor_make_shareable(rb_rational_new(numerator, denominator)); } /** @@ -264,10 +362,10 @@ parse_imaginary(const pm_imaginary_node_t *node) break; } default: - rb_bug("Unexpected numeric type on imaginary number %s\n", pm_node_type_to_str(PM_NODE_TYPE(node->numeric))); + rb_bug("Unexpected numeric type on imaginary number %s\n", pm_node_type(PM_NODE_TYPE(node->numeric))); } - return rb_complex_raw(INT2FIX(0), imaginary_part); + return RB_OBJ_SET_SHAREABLE(rb_complex_raw(INT2FIX(0), imaginary_part)); } static inline VALUE @@ -300,7 +398,7 @@ parse_string_encoded(const pm_node_t *node, const pm_string_t *string, rb_encodi } static inline VALUE -parse_static_literal_string(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, const pm_node_t *node, const pm_string_t *string) +parse_static_literal_string(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, const pm_string_t *string) { rb_encoding *encoding; @@ -318,11 +416,8 @@ parse_static_literal_string(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, rb_enc_str_coderange(value); if (ISEQ_COMPILE_DATA(iseq)->option->debug_frozen_string_literal || RTEST(ruby_debug)) { - int line_number = pm_node_line_number(scope_node->parser, node); - VALUE debug_info = rb_ary_new_from_args(2, rb_iseq_path(iseq), INT2FIX(line_number)); - value = rb_str_dup(value); - rb_ivar_set(value, id_debug_created_info, rb_ary_freeze(debug_info)); - rb_str_freeze(value); + int line_number = pm_node_line_number_cached(node, scope_node); + value = rb_ractor_make_shareable(rb_str_with_debug_created_info(value, rb_iseq_path(iseq), line_number)); } return value; @@ -366,7 +461,7 @@ parse_regexp_error(rb_iseq_t *iseq, int32_t line_number, const char *fmt, ...) } static VALUE -parse_regexp_string_part(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, const pm_node_t *node, const pm_string_t *unescaped, rb_encoding *implicit_regexp_encoding, rb_encoding *explicit_regexp_encoding) +parse_regexp_string_part(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, const pm_string_t *unescaped, rb_encoding *implicit_regexp_encoding, rb_encoding *explicit_regexp_encoding) { // If we were passed an explicit regexp encoding, then we need to double // check that it's okay here for this fragment of the string. @@ -388,12 +483,12 @@ parse_regexp_string_part(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, con VALUE string = rb_enc_str_new((const char *) pm_string_source(unescaped), pm_string_length(unescaped), encoding); VALUE error = rb_reg_check_preprocess(string); - if (error != Qnil) parse_regexp_error(iseq, pm_node_line_number(scope_node->parser, node), "%" PRIsVALUE, rb_obj_as_string(error)); + if (error != Qnil) parse_regexp_error(iseq, pm_node_line_number_cached(node, scope_node), "%" PRIsVALUE, rb_obj_as_string(error)); return string; } static VALUE -pm_static_literal_concat(rb_iseq_t *iseq, const pm_node_list_t *nodes, const pm_scope_node_t *scope_node, rb_encoding *implicit_regexp_encoding, rb_encoding *explicit_regexp_encoding, bool top) +pm_static_literal_concat(rb_iseq_t *iseq, const pm_node_list_t *nodes, pm_scope_node_t *scope_node, rb_encoding *implicit_regexp_encoding, rb_encoding *explicit_regexp_encoding, bool top) { VALUE current = Qnil; @@ -410,7 +505,7 @@ pm_static_literal_concat(rb_iseq_t *iseq, const pm_node_list_t *nodes, const pm_ else { string = parse_string_encoded(part, &((const pm_string_node_t *) part)->unescaped, scope_node->encoding); VALUE error = rb_reg_check_preprocess(string); - if (error != Qnil) parse_regexp_error(iseq, pm_node_line_number(scope_node->parser, part), "%" PRIsVALUE, rb_obj_as_string(error)); + if (error != Qnil) parse_regexp_error(iseq, pm_node_line_number_cached(part, scope_node), "%" PRIsVALUE, rb_obj_as_string(error)); } } else { @@ -505,7 +600,7 @@ parse_regexp_flags(const pm_node_t *node) static rb_encoding * parse_regexp_encoding(const pm_scope_node_t *scope_node, const pm_node_t *node) { - if (PM_NODE_FLAG_P(node, PM_REGULAR_EXPRESSION_FLAGS_ASCII_8BIT)) { + if (PM_NODE_FLAG_P(node, PM_REGULAR_EXPRESSION_FLAGS_FORCED_BINARY_ENCODING) || PM_NODE_FLAG_P(node, PM_REGULAR_EXPRESSION_FLAGS_ASCII_8BIT)) { return rb_ascii8bit_encoding(); } else if (PM_NODE_FLAG_P(node, PM_REGULAR_EXPRESSION_FLAGS_UTF_8)) { @@ -523,12 +618,12 @@ parse_regexp_encoding(const pm_scope_node_t *scope_node, const pm_node_t *node) } static VALUE -parse_regexp(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, const pm_node_t *node, VALUE string) +parse_regexp(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, VALUE string) { VALUE errinfo = rb_errinfo(); - int32_t line_number = pm_node_line_number(scope_node->parser, node); - VALUE regexp = rb_reg_compile(string, parse_regexp_flags(node), (const char *) pm_string_source(&scope_node->parser->filepath), line_number); + int32_t line_number = pm_node_line_number_cached(node, scope_node); + VALUE regexp = rb_reg_compile(string, parse_regexp_flags(node), (const char *) pm_string_source(pm_parser_filepath(scope_node->parser)), line_number); if (NIL_P(regexp)) { VALUE message = rb_attr_get(rb_errinfo(), idMesg); @@ -538,22 +633,22 @@ parse_regexp(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, const pm_node_t return Qnil; } - rb_obj_freeze(regexp); - return regexp; + return RB_OBJ_SET_SHAREABLE(rb_obj_freeze(regexp)); } static inline VALUE -parse_regexp_literal(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, const pm_node_t *node, const pm_string_t *unescaped) +parse_regexp_literal(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, const pm_string_t *unescaped) { rb_encoding *regexp_encoding = parse_regexp_encoding(scope_node, node); if (regexp_encoding == NULL) regexp_encoding = scope_node->encoding; VALUE string = rb_enc_str_new((const char *) pm_string_source(unescaped), pm_string_length(unescaped), regexp_encoding); + RB_OBJ_SET_SHAREABLE(string); return parse_regexp(iseq, scope_node, node, string); } static inline VALUE -parse_regexp_concat(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, const pm_node_t *node, const pm_node_list_t *parts) +parse_regexp_concat(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, const pm_node_list_t *parts) { rb_encoding *explicit_regexp_encoding = parse_regexp_encoding(scope_node, node); rb_encoding *implicit_regexp_encoding = explicit_regexp_encoding != NULL ? explicit_regexp_encoding : scope_node->encoding; @@ -565,7 +660,7 @@ parse_regexp_concat(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, const pm static void pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node); static int -pm_interpolated_node_compile(rb_iseq_t *iseq, const pm_node_list_t *parts, const pm_node_location_t *node_location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node, rb_encoding *implicit_regexp_encoding, rb_encoding *explicit_regexp_encoding) +pm_interpolated_node_compile(rb_iseq_t *iseq, const pm_node_list_t *parts, const pm_node_location_t *node_location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node, rb_encoding *implicit_regexp_encoding, rb_encoding *explicit_regexp_encoding, bool mutable_result, bool frozen_result) { int stack_size = 0; size_t parts_size = parts->size; @@ -594,7 +689,7 @@ pm_interpolated_node_compile(rb_iseq_t *iseq, const pm_node_list_t *parts, const } else { current_string = string_value; - if (index != 0) current_location = PM_NODE_END_LOCATION(scope_node->parser, part); + if (index != 0) current_location = PM_NODE_END_LOCATION(part); } } else { @@ -621,7 +716,7 @@ pm_interpolated_node_compile(rb_iseq_t *iseq, const pm_node_list_t *parts, const } else { current_string = string_value; - current_location = PM_NODE_START_LOCATION(scope_node->parser, part); + current_location = PM_NODE_START_LOCATION(part); } } else { @@ -632,7 +727,7 @@ pm_interpolated_node_compile(rb_iseq_t *iseq, const pm_node_list_t *parts, const if (explicit_regexp_encoding != NULL) { encoding = explicit_regexp_encoding; } - else if (scope_node->parser->encoding == PM_ENCODING_US_ASCII_ENTRY) { + else if (pm_parser_encoding_us_ascii(scope_node->parser)) { encoding = rb_ascii8bit_encoding(); } else { @@ -643,19 +738,31 @@ pm_interpolated_node_compile(rb_iseq_t *iseq, const pm_node_list_t *parts, const encoding = scope_node->encoding; } - current_string = rb_enc_str_new(NULL, 0, encoding); + if (parts_size == 1) { + current_string = rb_enc_str_new(NULL, 0, encoding); + } + } + + if (RTEST(current_string)) { + VALUE operand = rb_fstring(current_string); + PUSH_INSN1(ret, current_location, putobject, operand); + stack_size++; } - PUSH_INSN1(ret, current_location, putobject, rb_fstring(current_string)); PM_COMPILE_NOT_POPPED(part); - const pm_node_location_t current_location = PM_NODE_START_LOCATION(scope_node->parser, part); + const pm_node_location_t current_location = PM_NODE_START_LOCATION(part); PUSH_INSN(ret, current_location, dup); - PUSH_INSN1(ret, current_location, objtostring, new_callinfo(iseq, idTo_s, 0, VM_CALL_FCALL | VM_CALL_ARGS_SIMPLE, NULL, FALSE)); + + { + const struct rb_callinfo *callinfo = new_callinfo(iseq, idTo_s, 0, VM_CALL_FCALL | VM_CALL_ARGS_SIMPLE, NULL, FALSE); + PUSH_INSN1(ret, current_location, objtostring, callinfo); + } + PUSH_INSN(ret, current_location, anytostring); current_string = Qnil; - stack_size += 2; + stack_size++; } } } @@ -663,10 +770,15 @@ pm_interpolated_node_compile(rb_iseq_t *iseq, const pm_node_list_t *parts, const if (RTEST(current_string)) { current_string = rb_fstring(current_string); - if (stack_size == 0 && interpolated) { - PUSH_INSN1(ret, current_location, putstring, current_string); - } - else { + if (stack_size == 0) { + if (frozen_result) { + PUSH_INSN1(ret, current_location, putobject, current_string); + } else if (mutable_result || interpolated) { + PUSH_INSN1(ret, current_location, dupstring, current_string); + } else { + PUSH_INSN1(ret, current_location, dupchilledstring, current_string); + } + } else { PUSH_INSN1(ret, current_location, putobject, current_string); } @@ -687,7 +799,7 @@ pm_compile_regexp_dynamic(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_ rb_encoding *explicit_regexp_encoding = parse_regexp_encoding(scope_node, node); rb_encoding *implicit_regexp_encoding = explicit_regexp_encoding != NULL ? explicit_regexp_encoding : scope_node->encoding; - int length = pm_interpolated_node_compile(iseq, parts, node_location, ret, popped, scope_node, implicit_regexp_encoding, explicit_regexp_encoding); + int length = pm_interpolated_node_compile(iseq, parts, node_location, ret, popped, scope_node, implicit_regexp_encoding, explicit_regexp_encoding, false, false); PUSH_INSN2(ret, *node_location, toregexp, INT2FIX(parse_regexp_flags(node) & 0xFF), INT2FIX(length)); } @@ -714,9 +826,9 @@ static VALUE pm_static_literal_string(rb_iseq_t *iseq, VALUE string, int line_number) { if (ISEQ_COMPILE_DATA(iseq)->option->debug_frozen_string_literal || RTEST(ruby_debug)) { - VALUE debug_info = rb_ary_new_from_args(2, rb_iseq_path(iseq), INT2FIX(line_number)); - rb_ivar_set(string, id_debug_created_info, rb_ary_freeze(debug_info)); - return rb_str_freeze(string); + VALUE str = rb_str_with_debug_created_info(string, rb_iseq_path(iseq), line_number); + RB_OBJ_SET_SHAREABLE(str); + return str; } else { return rb_fstring(string); @@ -729,7 +841,7 @@ pm_static_literal_string(rb_iseq_t *iseq, VALUE string, int line_number) * literal values can be compiled into a literal array. */ static VALUE -pm_static_literal_value(rb_iseq_t *iseq, const pm_node_t *node, const pm_scope_node_t *scope_node) +pm_static_literal_value(rb_iseq_t *iseq, const pm_node_t *node, pm_scope_node_t *scope_node) { // Every node that comes into this function should already be marked as // static literal. If it's not, then we have a bug somewhere. @@ -745,7 +857,7 @@ pm_static_literal_value(rb_iseq_t *iseq, const pm_node_t *node, const pm_scope_n rb_ary_push(value, pm_static_literal_value(iseq, elements->nodes[index], scope_node)); } - OBJ_FREEZE(value); + RB_OBJ_SET_FROZEN_SHAREABLE(value); return value; } case PM_FALSE_NODE: @@ -764,11 +876,11 @@ pm_static_literal_value(rb_iseq_t *iseq, const pm_node_t *node, const pm_scope_n rb_ary_cat(array, pair, 2); } - VALUE value = rb_hash_new_with_size(elements->size); + VALUE value = rb_hash_alloc_fixed_size(Qfalse, elements->size); rb_hash_bulk_insert(RARRAY_LEN(array), RARRAY_CONST_PTR(array), value); + RB_GC_GUARD(array); - value = rb_obj_hide(value); - OBJ_FREEZE(value); + RB_OBJ_SET_FROZEN_SHAREABLE(value); return value; } case PM_IMAGINARY_NODE: @@ -785,7 +897,7 @@ pm_static_literal_value(rb_iseq_t *iseq, const pm_node_t *node, const pm_scope_n } case PM_INTERPOLATED_STRING_NODE: { VALUE string = pm_static_literal_concat(iseq, &((const pm_interpolated_string_node_t *) node)->parts, scope_node, NULL, NULL, false); - int line_number = pm_node_line_number(scope_node->parser, node); + int line_number = pm_node_line_number_cached(node, scope_node); return pm_static_literal_string(iseq, string, line_number); } case PM_INTERPOLATED_SYMBOL_NODE: { @@ -813,7 +925,7 @@ pm_static_literal_value(rb_iseq_t *iseq, const pm_node_t *node, const pm_scope_n return pm_source_file_value(cast, scope_node); } case PM_SOURCE_LINE_NODE: - return INT2FIX(pm_node_line_number(scope_node->parser, node)); + return INT2FIX(pm_node_line_number_cached(node, scope_node)); case PM_STRING_NODE: { const pm_string_node_t *cast = (const pm_string_node_t *) node; return parse_static_literal_string(iseq, scope_node, node, &cast->unescaped); @@ -823,7 +935,7 @@ pm_static_literal_value(rb_iseq_t *iseq, const pm_node_t *node, const pm_scope_n case PM_TRUE_NODE: return Qtrue; default: - rb_bug("Don't have a literal value for node type %s", pm_node_type_to_str(PM_NODE_TYPE(node))); + rb_bug("Don't have a literal value for node type %s", pm_node_type(PM_NODE_TYPE(node))); return Qfalse; } } @@ -832,10 +944,10 @@ pm_static_literal_value(rb_iseq_t *iseq, const pm_node_t *node, const pm_scope_n * A helper for converting a pm_location_t into a rb_code_location_t. */ static rb_code_location_t -pm_code_location(const pm_scope_node_t *scope_node, const pm_node_t *node) +pm_code_location(pm_scope_node_t *scope_node, const pm_node_t *node) { - const pm_line_column_t start_location = PM_NODE_START_LINE_COLUMN(scope_node->parser, node); - const pm_line_column_t end_location = PM_NODE_END_LINE_COLUMN(scope_node->parser, node); + const pm_line_column_t start_location = PM_NODE_START_LINE_COLUMN(node); + const pm_line_column_t end_location = PM_NODE_END_LINE_COLUMN(node); return (rb_code_location_t) { .beg_pos = { .lineno = start_location.line, .column = start_location.column }, @@ -852,31 +964,27 @@ pm_code_location(const pm_scope_node_t *scope_node, const pm_node_t *node) static void pm_compile_branch_condition(rb_iseq_t *iseq, LINK_ANCHOR *const ret, const pm_node_t *cond, - LABEL *then_label, LABEL *else_label, bool popped, pm_scope_node_t *scope_node); + LABEL *then_label, LABEL *else_label, pm_scope_node_t *scope_node); static void -pm_compile_logical(rb_iseq_t *iseq, LINK_ANCHOR *const ret, pm_node_t *cond, LABEL *then_label, LABEL *else_label, bool popped, pm_scope_node_t *scope_node) +pm_compile_logical(rb_iseq_t *iseq, LINK_ANCHOR *const ret, pm_node_t *cond, LABEL *then_label, LABEL *else_label, pm_scope_node_t *scope_node) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, cond); + const pm_node_location_t location = PM_NODE_START_LOCATION(cond); DECL_ANCHOR(seq); - INIT_ANCHOR(seq); LABEL *label = NEW_LABEL(location.line); if (!then_label) then_label = label; else if (!else_label) else_label = label; - pm_compile_branch_condition(iseq, seq, cond, then_label, else_label, popped, scope_node); + pm_compile_branch_condition(iseq, seq, cond, then_label, else_label, scope_node); if (LIST_INSN_SIZE_ONE(seq)) { INSN *insn = (INSN *) ELEM_FIRST_INSN(FIRST_ELEMENT(seq)); if (insn->insn_id == BIN(jump) && (LABEL *)(insn->operands[0]) == label) return; } - if (!label->refcnt) { - if (popped) PUSH_INSN(ret, location, putnil); - } - else { + if (label->refcnt) { PUSH_LABEL(seq, label); } @@ -887,11 +995,14 @@ pm_compile_logical(rb_iseq_t *iseq, LINK_ANCHOR *const ret, pm_node_t *cond, LAB static void pm_compile_flip_flop_bound(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) { - const pm_node_location_t location = { .line = ISEQ_BODY(iseq)->location.first_lineno, .node_id = -1 }; + const pm_node_location_t location = PM_NODE_START_LOCATION(node); if (PM_NODE_TYPE_P(node, PM_INTEGER_NODE)) { PM_COMPILE_NOT_POPPED(node); - PUSH_INSN1(ret, location, getglobal, ID2SYM(rb_intern("$."))); + + VALUE operand = ID2SYM(rb_intern("$.")); + PUSH_INSN1(ret, location, getglobal, operand); + PUSH_SEND(ret, location, idEq, INT2FIX(1)); if (popped) PUSH_INSN(ret, location, pop); } @@ -903,7 +1014,7 @@ pm_compile_flip_flop_bound(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR * static void pm_compile_flip_flop(const pm_flip_flop_node_t *flip_flop_node, LABEL *else_label, LABEL *then_label, rb_iseq_t *iseq, const int lineno, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) { - const pm_node_location_t location = { .line = ISEQ_BODY(iseq)->location.first_lineno, .node_id = -1 }; + const pm_node_location_t location = { .line = lineno, .node_id = -1 }; LABEL *lend = NEW_LABEL(location.line); int again = !(flip_flop_node->base.flags & PM_RANGE_FLAGS_EXCLUDE_END); @@ -945,22 +1056,22 @@ pm_compile_flip_flop(const pm_flip_flop_node_t *flip_flop_node, LABEL *else_labe static void pm_compile_defined_expr(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_location_t *node_location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node, bool in_condition); static void -pm_compile_branch_condition(rb_iseq_t *iseq, LINK_ANCHOR *const ret, const pm_node_t *cond, LABEL *then_label, LABEL *else_label, bool popped, pm_scope_node_t *scope_node) +pm_compile_branch_condition(rb_iseq_t *iseq, LINK_ANCHOR *const ret, const pm_node_t *cond, LABEL *then_label, LABEL *else_label, pm_scope_node_t *scope_node) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, cond); + const pm_node_location_t location = PM_NODE_START_LOCATION(cond); again: switch (PM_NODE_TYPE(cond)) { case PM_AND_NODE: { const pm_and_node_t *cast = (const pm_and_node_t *) cond; - pm_compile_logical(iseq, ret, cast->left, NULL, else_label, popped, scope_node); + pm_compile_logical(iseq, ret, cast->left, NULL, else_label, scope_node); cond = cast->right; goto again; } case PM_OR_NODE: { const pm_or_node_t *cast = (const pm_or_node_t *) cond; - pm_compile_logical(iseq, ret, cast->left, then_label, NULL, popped, scope_node); + pm_compile_logical(iseq, ret, cast->left, then_label, NULL, scope_node); cond = cast->right; goto again; @@ -981,28 +1092,28 @@ again: PUSH_INSNL(ret, location, jump, then_label); return; case PM_FLIP_FLOP_NODE: - pm_compile_flip_flop((const pm_flip_flop_node_t *) cond, else_label, then_label, iseq, location.line, ret, popped, scope_node); + pm_compile_flip_flop((const pm_flip_flop_node_t *) cond, else_label, then_label, iseq, location.line, ret, false, scope_node); return; case PM_DEFINED_NODE: { const pm_defined_node_t *cast = (const pm_defined_node_t *) cond; - pm_compile_defined_expr(iseq, cast->value, &location, ret, popped, scope_node, true); + pm_compile_defined_expr(iseq, cast->value, &location, ret, false, scope_node, true); break; } default: { DECL_ANCHOR(cond_seq); - INIT_ANCHOR(cond_seq); pm_compile_node(iseq, cond, cond_seq, false, scope_node); if (LIST_INSN_SIZE_ONE(cond_seq)) { - INSN *insn = (INSN *)ELEM_FIRST_INSN(FIRST_ELEMENT(cond_seq)); + INSN *insn = (INSN *) ELEM_FIRST_INSN(FIRST_ELEMENT(cond_seq)); + if (insn->insn_id == BIN(putobject)) { if (RTEST(insn->operands[0])) { - ADD_INSNL(ret, cond, jump, then_label); + PUSH_INSNL(ret, location, jump, then_label); // maybe unreachable return; } else { - ADD_INSNL(ret, cond, jump, else_label); + PUSH_INSNL(ret, location, jump, else_label); return; } } @@ -1021,7 +1132,7 @@ again: * Compile an if or unless node. */ static void -pm_compile_conditional(rb_iseq_t *iseq, const pm_node_location_t *node_location, pm_node_type_t type, const pm_node_t *node, const pm_statements_node_t *statements, const pm_node_t *consequent, const pm_node_t *predicate, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +pm_compile_conditional(rb_iseq_t *iseq, const pm_node_location_t *node_location, pm_node_type_t type, const pm_node_t *node, const pm_statements_node_t *statements, const pm_node_t *subsequent, const pm_node_t *predicate, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) { const pm_node_location_t location = *node_location; LABEL *then_label = NEW_LABEL(location.line); @@ -1029,8 +1140,7 @@ pm_compile_conditional(rb_iseq_t *iseq, const pm_node_location_t *node_location, LABEL *end_label = NULL; DECL_ANCHOR(cond_seq); - INIT_ANCHOR(cond_seq); - pm_compile_branch_condition(iseq, cond_seq, predicate, then_label, else_label, false, scope_node); + pm_compile_branch_condition(iseq, cond_seq, predicate, then_label, else_label, scope_node); PUSH_SEQ(ret, cond_seq); rb_code_location_t conditional_location = { 0 }; @@ -1045,7 +1155,6 @@ pm_compile_conditional(rb_iseq_t *iseq, const pm_node_location_t *node_location, PUSH_LABEL(ret, then_label); DECL_ANCHOR(then_seq); - INIT_ANCHOR(then_seq); if (statements != NULL) { pm_compile_node(iseq, (const pm_node_t *) statements, then_seq, popped, scope_node); @@ -1062,7 +1171,7 @@ pm_compile_conditional(rb_iseq_t *iseq, const pm_node_location_t *node_location, if (statements != NULL) { branch_location = pm_code_location(scope_node, (const pm_node_t *) statements); } else if (type == PM_IF_NODE) { - pm_line_column_t predicate_end = PM_NODE_END_LINE_COLUMN(scope_node->parser, predicate); + pm_line_column_t predicate_end = PM_NODE_END_LINE_COLUMN(predicate); branch_location = (rb_code_location_t) { .beg_pos = { .lineno = predicate_end.line, .column = predicate_end.column }, .end_pos = { .lineno = predicate_end.line, .column = predicate_end.column } @@ -1086,10 +1195,9 @@ pm_compile_conditional(rb_iseq_t *iseq, const pm_node_location_t *node_location, PUSH_LABEL(ret, else_label); DECL_ANCHOR(else_seq); - INIT_ANCHOR(else_seq); - if (consequent != NULL) { - pm_compile_node(iseq, consequent, else_seq, popped, scope_node); + if (subsequent != NULL) { + pm_compile_node(iseq, subsequent, else_seq, popped, scope_node); } else if (!popped) { PUSH_SYNTHETIC_PUTNIL(else_seq, iseq); @@ -1099,13 +1207,13 @@ pm_compile_conditional(rb_iseq_t *iseq, const pm_node_location_t *node_location, if (then_label->refcnt && PM_BRANCH_COVERAGE_P(iseq)) { rb_code_location_t branch_location; - if (consequent == NULL) { + if (subsequent == NULL) { branch_location = conditional_location; - } else if (PM_NODE_TYPE_P(consequent, PM_ELSE_NODE)) { - const pm_else_node_t *else_node = (const pm_else_node_t *) consequent; + } else if (PM_NODE_TYPE_P(subsequent, PM_ELSE_NODE)) { + const pm_else_node_t *else_node = (const pm_else_node_t *) subsequent; branch_location = pm_code_location(scope_node, else_node->statements != NULL ? ((const pm_node_t *) else_node->statements) : (const pm_node_t *) else_node); } else { - branch_location = pm_code_location(scope_node, (const pm_node_t *) consequent); + branch_location = pm_code_location(scope_node, (const pm_node_t *) subsequent); } add_trace_branch_coverage(iseq, ret, &branch_location, branch_location.beg_pos.column, 1, type == PM_IF_NODE ? "else" : "then", branches); @@ -1180,10 +1288,10 @@ pm_compile_loop(rb_iseq_t *iseq, const pm_node_location_t *node_location, pm_nod PUSH_LABEL(ret, next_label); if (type == PM_WHILE_NODE) { - pm_compile_branch_condition(iseq, ret, predicate, redo_label, end_label, popped, scope_node); + pm_compile_branch_condition(iseq, ret, predicate, redo_label, end_label, scope_node); } else if (type == PM_UNTIL_NODE) { - pm_compile_branch_condition(iseq, ret, predicate, end_label, redo_label, popped, scope_node); + pm_compile_branch_condition(iseq, ret, predicate, end_label, redo_label, scope_node); } PUSH_LABEL(ret, end_label); @@ -1212,14 +1320,15 @@ static pm_local_index_t pm_lookup_local_index(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, pm_constant_id_t constant_id, int start_depth) { pm_local_index_t lindex = { 0 }; - st_data_t local_index; + int local_index; int level; for (level = 0; level < start_depth; level++) { scope_node = scope_node->previous; } - while (!st_lookup(scope_node->index_lookup_table, constant_id, &local_index)) { + while (!pm_index_lookup_table_lookup(&scope_node->index_lookup_table, constant_id, &local_index)) + { level++; if (scope_node->previous) { @@ -1243,12 +1352,10 @@ pm_lookup_local_index(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, pm_con // We add a constants mapping on the scope_node which is a mapping from // these constant_id indexes to the CRuby IDs that they represent. // This helper method allows easy access to those IDs -static ID +static inline ID pm_constant_id_lookup(const pm_scope_node_t *scope_node, pm_constant_id_t constant_id) { - if (constant_id < 1 || constant_id > scope_node->parser->constant_pool.size) { - rb_bug("constant_id out of range: %u", (unsigned int)constant_id); - } + RUBY_ASSERT(constant_id >= 1 && constant_id <= pm_parser_constants_size(scope_node->parser)); return scope_node->constants[constant_id - 1]; } @@ -1257,7 +1364,7 @@ pm_new_child_iseq(rb_iseq_t *iseq, pm_scope_node_t *node, VALUE name, const rb_i { debugs("[new_child_iseq]> ---------------------------------------\n"); int isolated_depth = ISEQ_COMPILE_DATA(iseq)->isolated_depth; - rb_iseq_t *ret_iseq = pm_iseq_new_with_opt(node, name, + rb_iseq_t *ret_iseq = pm_iseq_build(node, name, rb_iseq_path(iseq), rb_iseq_realpath(iseq), line_no, parent, isolated_depth ? isolated_depth + 1 : 0, @@ -1267,15 +1374,36 @@ pm_new_child_iseq(rb_iseq_t *iseq, pm_scope_node_t *node, VALUE name, const rb_i } static int +pm_cpath_const_p(const pm_node_t *node) +{ + switch (PM_NODE_TYPE(node)) { + case PM_CONSTANT_READ_NODE: + return TRUE; + case PM_CONSTANT_PATH_NODE: + { + const pm_node_t *parent = ((const pm_constant_path_node_t *) node)->parent; + if (!parent) return TRUE; /* ::Foo */ + return pm_cpath_const_p(parent); + } + default: + return FALSE; + } +} + +static int pm_compile_class_path(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_location_t *node_location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) { if (PM_NODE_TYPE_P(node, PM_CONSTANT_PATH_NODE)) { const pm_node_t *parent = ((const pm_constant_path_node_t *) node)->parent; if (parent) { - /* Bar::Foo */ + /* Bar::Foo or expr::Foo */ PM_COMPILE(parent); - return VM_DEFINECLASS_FLAG_SCOPED; + int flags = VM_DEFINECLASS_FLAG_SCOPED; + if (!pm_cpath_const_p(parent)) { + flags |= VM_DEFINECLASS_FLAG_DYNAMIC_CREF; + } + return flags; } else { /* toplevel class ::Foo */ @@ -1345,15 +1473,17 @@ pm_compile_call_and_or_write_node(rb_iseq_t *iseq, bool and_node, const pm_node_ if (lskip && !popped) PUSH_LABEL(ret, lskip); } +static void pm_compile_shareable_constant_value(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_flags_t shareability, VALUE path, LINK_ANCHOR *const ret, pm_scope_node_t *scope_node, bool top); + /** * This function compiles a hash onto the stack. It is used to compile hash * literals and keyword arguments. It is assumed that if we get here that the * contents of the hash are not popped. */ static void -pm_compile_hash_elements(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_list_t *elements, bool argument, LINK_ANCHOR *const ret, pm_scope_node_t *scope_node) +pm_compile_hash_elements(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_list_t *elements, const pm_node_flags_t shareability, VALUE path, bool argument, LINK_ANCHOR *const ret, pm_scope_node_t *scope_node) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); // If this element is not popped, then we need to create the hash on the // stack. Neighboring plain assoc nodes should be grouped together (either @@ -1372,7 +1502,6 @@ pm_compile_hash_elements(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_l bool static_literal = false; DECL_ANCHOR(anchor); - INIT_ANCHOR(anchor); // Convert pushed elements to a hash, and merge if needed. #define FLUSH_CHUNK \ @@ -1398,12 +1527,18 @@ pm_compile_hash_elements(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_l switch (PM_NODE_TYPE(element)) { case PM_ASSOC_NODE: { // Pre-allocation check (this branch can be omitted). - if (PM_NODE_FLAG_P(element, PM_NODE_FLAG_STATIC_LITERAL) && !static_literal && ((index + min_tmp_hash_length) < elements->size)) { + if ( + (shareability == 0) && + PM_NODE_FLAG_P(element, PM_NODE_FLAG_STATIC_LITERAL) && ( + (!static_literal && ((index + min_tmp_hash_length) < elements->size)) || + (first_chunk && stack_length == 0) + ) + ) { // Count the elements that are statically-known. size_t count = 1; while (index + count < elements->size && PM_NODE_FLAG_P(elements->nodes[index + count], PM_NODE_FLAG_STATIC_LITERAL)) count++; - if (count >= min_tmp_hash_length) { + if ((first_chunk && stack_length == 0) || count >= min_tmp_hash_length) { // The subsequence of elements in this hash is long enough // to merit its own hash. VALUE ary = rb_ary_hidden_new(count); @@ -1419,11 +1554,12 @@ pm_compile_hash_elements(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_l rb_ary_cat(ary, elem, 2); } + index --; - VALUE hash = rb_hash_new_with_size(RARRAY_LEN(ary) / 2); + VALUE hash = rb_hash_alloc_fixed_size(Qfalse, RARRAY_LEN(ary) / 2); rb_hash_bulk_insert(RARRAY_LEN(ary), RARRAY_CONST_PTR(ary), hash); - hash = rb_obj_hide(hash); - OBJ_FREEZE(hash); + RB_GC_GUARD(ary); + RB_OBJ_SET_FROZEN_SHAREABLE(hash); // Emit optimized code. FLUSH_CHUNK; @@ -1450,7 +1586,15 @@ pm_compile_hash_elements(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_l // If this is a plain assoc node, then we can compile it directly // and then add the total number of values on the stack. - pm_compile_node(iseq, element, anchor, false, scope_node); + if (shareability == 0) { + pm_compile_node(iseq, element, anchor, false, scope_node); + } + else { + const pm_assoc_node_t *assoc = (const pm_assoc_node_t *) element; + pm_compile_shareable_constant_value(iseq, assoc->key, shareability, path, ret, scope_node, false); + pm_compile_shareable_constant_value(iseq, assoc->value, shareability, path, ret, scope_node, false); + } + if ((stack_length += 2) >= max_stack_length) FLUSH_CHUNK; break; } @@ -1493,7 +1637,12 @@ pm_compile_hash_elements(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_l // only done for method calls and not for literal hashes, // because literal hashes should always result in a new // hash. - PM_COMPILE_NOT_POPPED(element); + if (shareability == 0) { + PM_COMPILE_NOT_POPPED(element); + } + else { + pm_compile_shareable_constant_value(iseq, element, shareability, path, ret, scope_node, false); + } } else { // There is more than one keyword argument, or this is not a @@ -1509,7 +1658,13 @@ pm_compile_hash_elements(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_l PUSH_INSN(ret, location, swap); } - PM_COMPILE_NOT_POPPED(element); + if (shareability == 0) { + PM_COMPILE_NOT_POPPED(element); + } + else { + pm_compile_shareable_constant_value(iseq, element, shareability, path, ret, scope_node, false); + } + PUSH_SEND(ret, location, id_core_hash_merge_kwd, INT2FIX(2)); } } @@ -1528,9 +1683,13 @@ pm_compile_hash_elements(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_l #undef FLUSH_CHUNK } +#define SPLATARRAY_FALSE 0 +#define SPLATARRAY_TRUE 1 +#define DUP_SINGLE_KW_SPLAT 2 + // This is details. Users should call pm_setup_args() instead. static int -pm_setup_args_core(const pm_arguments_node_t *arguments_node, const pm_node_t *block, int *flags, const bool has_regular_blockarg, struct rb_callinfo_kwarg **kw_arg, rb_iseq_t *iseq, LINK_ANCHOR *const ret, pm_scope_node_t *scope_node, const pm_node_location_t *node_location) +pm_setup_args_core(const pm_arguments_node_t *arguments_node, const pm_node_t *block, int *flags, const bool has_regular_blockarg, struct rb_callinfo_kwarg **kw_arg, int *dup_rest, rb_iseq_t *iseq, LINK_ANCHOR *const ret, pm_scope_node_t *scope_node, const pm_node_location_t *node_location) { const pm_node_location_t location = *node_location; @@ -1563,14 +1722,23 @@ pm_setup_args_core(const pm_arguments_node_t *arguments_node, const pm_node_t *b *flags |= VM_CALL_KW_SPLAT; has_keyword_splat = true; - if (elements->size > 1) { + if (elements->size > 1 || !(elements->size == 1 && PM_NODE_TYPE_P(elements->nodes[0], PM_ASSOC_SPLAT_NODE))) { // A new hash will be created for the keyword arguments // in this case, so mark the method as passing mutable // keyword splat. *flags |= VM_CALL_KW_SPLAT_MUT; + pm_compile_hash_elements(iseq, argument, elements, 0, Qundef, true, ret, scope_node); + } + else if (*dup_rest & DUP_SINGLE_KW_SPLAT) { + *flags |= VM_CALL_KW_SPLAT_MUT; + PUSH_INSN1(ret, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); + PUSH_INSN1(ret, location, newhash, INT2FIX(0)); + pm_compile_hash_elements(iseq, argument, elements, 0, Qundef, true, ret, scope_node); + PUSH_SEND(ret, location, id_core_hash_merge_kwd, INT2FIX(2)); + } + else { + pm_compile_hash_elements(iseq, argument, elements, 0, Qundef, true, ret, scope_node); } - - pm_compile_hash_elements(iseq, argument, elements, true, ret, scope_node); } else { // We need to first figure out if all elements of the @@ -1676,8 +1844,8 @@ pm_setup_args_core(const pm_arguments_node_t *arguments_node, const pm_node_t *b // foo(a, *b, c) // ^^ if (index + 1 < arguments->size || has_regular_blockarg) { - PUSH_INSN1(ret, location, splatarray, Qtrue); - *flags |= VM_CALL_ARGS_SPLAT_MUT; + PUSH_INSN1(ret, location, splatarray, (*dup_rest & SPLATARRAY_TRUE) ? Qtrue : Qfalse); + if (*dup_rest & SPLATARRAY_TRUE) *dup_rest &= ~SPLATARRAY_TRUE; } // If this is the first spalt array seen and it's the last // parameter, we don't want splatarray to dup it. @@ -1703,7 +1871,9 @@ pm_setup_args_core(const pm_arguments_node_t *arguments_node, const pm_node_t *b break; } - case PM_FORWARDING_ARGUMENTS_NODE: { + case PM_FORWARDING_ARGUMENTS_NODE: { // not counted in argc return value + iseq_set_use_block(ISEQ_BODY(iseq)->local_iseq); + if (ISEQ_BODY(ISEQ_BODY(iseq)->local_iseq)->param.flags.forwardable) { *flags |= VM_CALL_FORWARDING; @@ -1713,9 +1883,14 @@ pm_setup_args_core(const pm_arguments_node_t *arguments_node, const pm_node_t *b break; } - orig_argc += 2; + if (has_splat) { + // If we already have a splat, we're concatenating to existing array + orig_argc += 1; + } else { + orig_argc += 2; + } - *flags |= VM_CALL_ARGS_SPLAT | VM_CALL_ARGS_SPLAT_MUT | VM_CALL_ARGS_BLOCKARG | VM_CALL_KW_SPLAT; + *flags |= VM_CALL_ARGS_SPLAT | VM_CALL_ARGS_BLOCKARG | VM_CALL_KW_SPLAT; // Forwarding arguments nodes are treated as foo(*, **, &) // So foo(...) equals foo(*, **, &) and as such the local @@ -1724,7 +1899,13 @@ pm_setup_args_core(const pm_arguments_node_t *arguments_node, const pm_node_t *b // Push the * pm_local_index_t mult_local = pm_lookup_local_index(iseq, scope_node, PM_CONSTANT_MULT, 0); PUSH_GETLOCAL(ret, location, mult_local.index, mult_local.level); - PUSH_INSN1(ret, location, splatarray, Qtrue); + + if (has_splat) { + // If we already have a splat, we need to concatenate arrays + PUSH_INSN(ret, location, concattoarray); + } else { + PUSH_INSN1(ret, location, splatarray, Qfalse); + } // Push the ** pm_local_index_t pow_local = pm_lookup_local_index(iseq, scope_node, PM_CONSTANT_POW, 0); @@ -1733,7 +1914,6 @@ pm_setup_args_core(const pm_arguments_node_t *arguments_node, const pm_node_t *b // Push the & pm_local_index_t and_local = pm_lookup_local_index(iseq, scope_node, PM_CONSTANT_AND, 0); PUSH_INSN2(ret, location, getblockparamproxy, INT2FIX(and_local.index + VM_ENV_DATA_SIZE - 1), INT2FIX(and_local.level)); - PUSH_INSN(ret, location, splatkw); break; } @@ -1756,6 +1936,10 @@ pm_setup_args_core(const pm_arguments_node_t *arguments_node, const pm_node_t *b // foo(*a, b, c: :d) // foo(*a, b, **c) // + // If the next node is a forwarding argument: + // + // foo(*a, b, ...) + // // If the next node is NULL (we have hit the end): // // foo(*a, b) @@ -1778,6 +1962,10 @@ pm_setup_args_core(const pm_arguments_node_t *arguments_node, const pm_node_t *b PUSH_INSN(ret, location, concatarray); break; } + case PM_FORWARDING_ARGUMENTS_NODE: { + PUSH_INSN1(ret, location, pushtoarray, INT2FIX(post_splat_counter)); + break; + } default: break; } @@ -1796,17 +1984,121 @@ pm_setup_args_core(const pm_arguments_node_t *arguments_node, const pm_node_t *b return orig_argc; } -// Compile the argument parts of a call +/** + * True if the given kind of node could potentially mutate the array that is + * being splatted in a set of call arguments. + */ +static inline bool +pm_setup_args_dup_rest_p(const pm_node_t *node) +{ + switch (PM_NODE_TYPE(node)) { + case PM_BACK_REFERENCE_READ_NODE: + case PM_CLASS_VARIABLE_READ_NODE: + case PM_CONSTANT_READ_NODE: + case PM_FALSE_NODE: + case PM_FLOAT_NODE: + case PM_GLOBAL_VARIABLE_READ_NODE: + case PM_IMAGINARY_NODE: + case PM_INSTANCE_VARIABLE_READ_NODE: + case PM_INTEGER_NODE: + case PM_LAMBDA_NODE: + case PM_LOCAL_VARIABLE_READ_NODE: + case PM_NIL_NODE: + case PM_NUMBERED_REFERENCE_READ_NODE: + case PM_RATIONAL_NODE: + case PM_REGULAR_EXPRESSION_NODE: + case PM_SELF_NODE: + case PM_STRING_NODE: + case PM_SYMBOL_NODE: + case PM_TRUE_NODE: + return false; + case PM_CONSTANT_PATH_NODE: { + const pm_constant_path_node_t *cast = (const pm_constant_path_node_t *) node; + if (cast->parent != NULL) { + return pm_setup_args_dup_rest_p(cast->parent); + } + return false; + } + case PM_IMPLICIT_NODE: + return pm_setup_args_dup_rest_p(((const pm_implicit_node_t *) node)->value); + case PM_ARRAY_NODE: { + const pm_array_node_t *cast = (const pm_array_node_t *) node; + for (size_t index = 0; index < cast->elements.size; index++) { + if (pm_setup_args_dup_rest_p(cast->elements.nodes[index])) { + return true; + } + } + return false; + } + default: + return true; + } +} + +/** + * Compile the argument parts of a call. + */ static int pm_setup_args(const pm_arguments_node_t *arguments_node, const pm_node_t *block, int *flags, struct rb_callinfo_kwarg **kw_arg, rb_iseq_t *iseq, LINK_ANCHOR *const ret, pm_scope_node_t *scope_node, const pm_node_location_t *node_location) { + int dup_rest = SPLATARRAY_TRUE; + + const pm_node_list_t *arguments; + size_t arguments_size; + + // Calls like foo(1, *f, **hash) that use splat and kwsplat could be + // eligible for eliding duping the rest array (dup_reset=false). + if ( + arguments_node != NULL && + (arguments = &arguments_node->arguments, arguments_size = arguments->size) >= 2 && + PM_NODE_FLAG_P(arguments_node, PM_ARGUMENTS_NODE_FLAGS_CONTAINS_SPLAT) && + !PM_NODE_FLAG_P(arguments_node, PM_ARGUMENTS_NODE_FLAGS_CONTAINS_MULTIPLE_SPLATS) && + PM_NODE_TYPE_P(arguments->nodes[arguments_size - 1], PM_KEYWORD_HASH_NODE) + ) { + // Start by assuming that dup_rest=false, then check each element of the + // hash to ensure we don't need to flip it back to true (in case one of + // the elements could potentially mutate the array). + dup_rest = SPLATARRAY_FALSE; + + const pm_keyword_hash_node_t *keyword_hash = (const pm_keyword_hash_node_t *) arguments->nodes[arguments_size - 1]; + const pm_node_list_t *elements = &keyword_hash->elements; + + for (size_t index = 0; dup_rest == SPLATARRAY_FALSE && index < elements->size; index++) { + const pm_node_t *element = elements->nodes[index]; + + switch (PM_NODE_TYPE(element)) { + case PM_ASSOC_NODE: { + const pm_assoc_node_t *assoc = (const pm_assoc_node_t *) element; + if (pm_setup_args_dup_rest_p(assoc->key) || pm_setup_args_dup_rest_p(assoc->value)) dup_rest = SPLATARRAY_TRUE; + break; + } + case PM_ASSOC_SPLAT_NODE: { + const pm_assoc_splat_node_t *assoc = (const pm_assoc_splat_node_t *) element; + if (assoc->value != NULL && pm_setup_args_dup_rest_p(assoc->value)) dup_rest = SPLATARRAY_TRUE; + break; + } + default: + break; + } + } + } + + int initial_dup_rest = dup_rest; + int argc; + if (block && PM_NODE_TYPE_P(block, PM_BLOCK_ARGUMENT_NODE)) { // We compile the `&block_arg` expression first and stitch it later // since the nature of the expression influences whether splat should // duplicate the array. bool regular_block_arg = true; + const pm_node_t *block_expr = ((const pm_block_argument_node_t *)block)->expression; + + if (block_expr && pm_setup_args_dup_rest_p(block_expr)) { + dup_rest = SPLATARRAY_TRUE | DUP_SINGLE_KW_SPLAT; + initial_dup_rest = dup_rest; + } + DECL_ANCHOR(block_arg); - INIT_ANCHOR(block_arg); pm_compile_node(iseq, block, block_arg, false, scope_node); *flags |= VM_CALL_ARGS_BLOCKARG; @@ -1818,19 +2110,31 @@ pm_setup_args(const pm_arguments_node_t *arguments_node, const pm_node_t *block, if (iobj->insn_id == BIN(getblockparam)) { iobj->insn_id = BIN(getblockparamproxy); } + // Allow splat without duplication for simple one-instruction - // block arguments like `&arg`. It is known that this optimization - // can be too aggressive in some cases. See [Bug #16504]. + // block arguments like `&arg`. It is known that this + // optimization can be too aggressive in some cases. See + // [Bug #16504]. regular_block_arg = false; } } - int argc = pm_setup_args_core(arguments_node, block, flags, regular_block_arg, kw_arg, iseq, ret, scope_node, node_location); + argc = pm_setup_args_core(arguments_node, block, flags, regular_block_arg, kw_arg, &dup_rest, iseq, ret, scope_node, node_location); PUSH_SEQ(ret, block_arg); - return argc; + } + else { + argc = pm_setup_args_core(arguments_node, block, flags, false, kw_arg, &dup_rest, iseq, ret, scope_node, node_location); + } + + // If the dup_rest flag was consumed while compiling the arguments (which + // effectively means we found the splat node), then it would have changed + // during the call to pm_setup_args_core. In this case, we want to add the + // VM_CALL_ARGS_SPLAT_MUT flag. + if (*flags & VM_CALL_ARGS_SPLAT && dup_rest != initial_dup_rest) { + *flags |= VM_CALL_ARGS_SPLAT_MUT; } - return pm_setup_args_core(arguments_node, block, flags, false, kw_arg, iseq, ret, scope_node, node_location); + return argc; } /** @@ -1854,7 +2158,7 @@ pm_compile_index_operator_write_node(rb_iseq_t *iseq, const pm_index_operator_wr int boff = (node->block == NULL ? 0 : 1); int flag = PM_NODE_TYPE_P(node->receiver, PM_SELF_NODE) ? VM_CALL_FCALL : 0; struct rb_callinfo_kwarg *keywords = NULL; - int argc = pm_setup_args(node->arguments, node->block, &flag, &keywords, iseq, ret, scope_node, node_location); + int argc = pm_setup_args(node->arguments, (const pm_node_t *) node->block, &flag, &keywords, iseq, ret, scope_node, node_location); if ((argc > 0 || boff) && (flag & VM_CALL_KW_SPLAT)) { if (boff) { @@ -1961,7 +2265,7 @@ pm_compile_index_operator_write_node(rb_iseq_t *iseq, const pm_index_operator_wr * []= method. */ static void -pm_compile_index_control_flow_write_node(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_t *receiver, const pm_arguments_node_t *arguments, const pm_node_t *block, const pm_node_t *value, const pm_node_location_t *node_location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +pm_compile_index_control_flow_write_node(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_t *receiver, const pm_arguments_node_t *arguments, const pm_block_argument_node_t *block, const pm_node_t *value, const pm_node_location_t *node_location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) { const pm_node_location_t location = *node_location; if (!popped) PUSH_INSN(ret, location, putnil); @@ -1970,7 +2274,7 @@ pm_compile_index_control_flow_write_node(rb_iseq_t *iseq, const pm_node_t *node, int boff = (block == NULL ? 0 : 1); int flag = PM_NODE_TYPE_P(receiver, PM_SELF_NODE) ? VM_CALL_FCALL : 0; struct rb_callinfo_kwarg *keywords = NULL; - int argc = pm_setup_args(arguments, block, &flag, &keywords, iseq, ret, scope_node, node_location); + int argc = pm_setup_args(arguments, (const pm_node_t *) block, &flag, &keywords, iseq, ret, scope_node, node_location); if ((argc > 0 || boff) && (flag & VM_CALL_KW_SPLAT)) { if (boff) { @@ -2095,7 +2399,7 @@ pm_compile_index_control_flow_write_node(rb_iseq_t *iseq, const pm_node_t *node, // A forward declaration because this is the recursive function that handles // compiling a pattern. It can be reentered by nesting patterns, as in the case // of arrays or hashes. -static int pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, LINK_ANCHOR *const ret, LABEL *matched_label, LABEL *unmatched_label, bool in_single_pattern, bool in_alternation_pattern, bool use_deconstructed_cache, unsigned int base_index); +static int pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, LINK_ANCHOR *const ret, LABEL *matched_label, LABEL *unmatched_label, bool in_single_pattern, bool use_deconstructed_cache, unsigned int base_index); /** * This function generates the code to set up the error string and error_p @@ -2104,7 +2408,7 @@ static int pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, cons static int pm_compile_pattern_generic_error(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, LINK_ANCHOR *const ret, VALUE message, unsigned int base_index) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); LABEL *match_succeeded_label = NEW_LABEL(location.line); PUSH_INSN(ret, location, dup); @@ -2134,7 +2438,7 @@ pm_compile_pattern_generic_error(rb_iseq_t *iseq, pm_scope_node_t *scope_node, c static int pm_compile_pattern_length_error(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, LINK_ANCHOR *const ret, VALUE message, VALUE length, unsigned int base_index) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); LABEL *match_succeeded_label = NEW_LABEL(location.line); PUSH_INSN(ret, location, dup); @@ -2167,14 +2471,16 @@ pm_compile_pattern_length_error(rb_iseq_t *iseq, pm_scope_node_t *scope_node, co static int pm_compile_pattern_eqq_error(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, LINK_ANCHOR *const ret, unsigned int base_index) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); LABEL *match_succeeded_label = NEW_LABEL(location.line); PUSH_INSN(ret, location, dup); PUSH_INSNL(ret, location, branchif, match_succeeded_label); - PUSH_INSN1(ret, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); - PUSH_INSN1(ret, location, putobject, rb_fstring_lit("%p === %p does not return true")); + + VALUE operand = rb_fstring_lit("%p === %p does not return true"); + PUSH_INSN1(ret, location, putobject, operand); + PUSH_INSN1(ret, location, topn, INT2FIX(3)); PUSH_INSN1(ret, location, topn, INT2FIX(5)); PUSH_SEND(ret, location, id_core_sprintf, INT2FIX(3)); @@ -2199,10 +2505,10 @@ pm_compile_pattern_eqq_error(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const * label. */ static int -pm_compile_pattern_match(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, LINK_ANCHOR *const ret, LABEL *unmatched_label, bool in_single_pattern, bool in_alternation_pattern, bool use_deconstructed_cache, unsigned int base_index) +pm_compile_pattern_match(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, LINK_ANCHOR *const ret, LABEL *unmatched_label, bool in_single_pattern, bool use_deconstructed_cache, unsigned int base_index) { - LABEL *matched_label = NEW_LABEL(pm_node_line_number(scope_node->parser, node)); - CHECK(pm_compile_pattern(iseq, scope_node, node, ret, matched_label, unmatched_label, in_single_pattern, in_alternation_pattern, use_deconstructed_cache, base_index)); + LABEL *matched_label = NEW_LABEL(pm_node_line_number_cached(node, scope_node)); + CHECK(pm_compile_pattern(iseq, scope_node, node, ret, matched_label, unmatched_label, in_single_pattern, use_deconstructed_cache, base_index)); PUSH_LABEL(ret, matched_label); return COMPILE_OK; } @@ -2215,7 +2521,7 @@ pm_compile_pattern_match(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_ static int pm_compile_pattern_deconstruct(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, LINK_ANCHOR *const ret, LABEL *deconstruct_label, LABEL *match_failed_label, LABEL *deconstructed_label, LABEL *type_error_label, bool in_single_pattern, bool use_deconstructed_cache, unsigned int base_index) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); if (use_deconstructed_cache) { PUSH_INSN1(ret, location, topn, INT2FIX(base_index + PM_PATTERN_BASE_INDEX_OFFSET_DECONSTRUCTED_CACHE)); @@ -2234,7 +2540,9 @@ pm_compile_pattern_deconstruct(rb_iseq_t *iseq, pm_scope_node_t *scope_node, con PUSH_LABEL(ret, deconstruct_label); PUSH_INSN(ret, location, dup); - PUSH_INSN1(ret, location, putobject, ID2SYM(rb_intern("deconstruct"))); + + VALUE operand = ID2SYM(rb_intern("deconstruct")); + PUSH_INSN1(ret, location, putobject, operand); PUSH_SEND(ret, location, idRespond_to, INT2FIX(1)); if (use_deconstructed_cache) { @@ -2267,7 +2575,7 @@ pm_compile_pattern_deconstruct(rb_iseq_t *iseq, pm_scope_node_t *scope_node, con static int pm_compile_pattern_constant(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, LINK_ANCHOR *const ret, LABEL *match_failed_label, bool in_single_pattern, unsigned int base_index) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); PUSH_INSN(ret, location, dup); PM_COMPILE_NOT_POPPED(node); @@ -2288,9 +2596,9 @@ pm_compile_pattern_constant(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const * responsible for compiling in those error raising instructions. */ static void -pm_compile_pattern_error_handler(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, const pm_node_t *node, LINK_ANCHOR *const ret, LABEL *done_label, bool popped) +pm_compile_pattern_error_handler(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, LINK_ANCHOR *const ret, LABEL *done_label, bool popped) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); LABEL *key_error_label = NEW_LABEL(location.line); LABEL *cleanup_label = NEW_LABEL(location.line); @@ -2306,7 +2614,12 @@ pm_compile_pattern_error_handler(rb_iseq_t *iseq, const pm_scope_node_t *scope_n PUSH_INSN1(ret, location, putobject, rb_eNoMatchingPatternError); PUSH_INSN1(ret, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); - PUSH_INSN1(ret, location, putobject, rb_fstring_lit("%p: %s")); + + { + VALUE operand = rb_fstring_lit("%p: %s"); + PUSH_INSN1(ret, location, putobject, operand); + } + PUSH_INSN1(ret, location, topn, INT2FIX(4)); PUSH_INSN1(ret, location, topn, INT2FIX(PM_PATTERN_BASE_INDEX_OFFSET_ERROR_STRING + 6)); PUSH_SEND(ret, location, id_core_sprintf, INT2FIX(3)); @@ -2316,7 +2629,12 @@ pm_compile_pattern_error_handler(rb_iseq_t *iseq, const pm_scope_node_t *scope_n PUSH_LABEL(ret, key_error_label); PUSH_INSN1(ret, location, putobject, rb_eNoMatchingPatternKeyError); PUSH_INSN1(ret, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); - PUSH_INSN1(ret, location, putobject, rb_fstring_lit("%p: %s")); + + { + VALUE operand = rb_fstring_lit("%p: %s"); + PUSH_INSN1(ret, location, putobject, operand); + } + PUSH_INSN1(ret, location, topn, INT2FIX(4)); PUSH_INSN1(ret, location, topn, INT2FIX(PM_PATTERN_BASE_INDEX_OFFSET_ERROR_STRING + 6)); PUSH_SEND(ret, location, id_core_sprintf, INT2FIX(3)); @@ -2337,9 +2655,9 @@ pm_compile_pattern_error_handler(rb_iseq_t *iseq, const pm_scope_node_t *scope_n * Compile a pattern matching expression. */ static int -pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, LINK_ANCHOR *const ret, LABEL *matched_label, LABEL *unmatched_label, bool in_single_pattern, bool in_alternation_pattern, bool use_deconstructed_cache, unsigned int base_index) +pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t *node, LINK_ANCHOR *const ret, LABEL *matched_label, LABEL *unmatched_label, bool in_single_pattern, bool use_deconstructed_cache, unsigned int base_index) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); switch (PM_NODE_TYPE(node)) { case PM_ARRAY_PATTERN_NODE: { @@ -2397,7 +2715,7 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t PUSH_INSN(ret, location, dup); PUSH_INSN1(ret, location, putobject, INT2FIX(index)); PUSH_SEND(ret, location, idAREF, INT2FIX(1)); - CHECK(pm_compile_pattern_match(iseq, scope_node, required, ret, match_failed_label, in_single_pattern, in_alternation_pattern, false, base_index + 1)); + CHECK(pm_compile_pattern_match(iseq, scope_node, required, ret, match_failed_label, in_single_pattern, false, base_index + 1)); } if (cast->rest != NULL) { @@ -2410,7 +2728,7 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t PUSH_SEND(ret, location, idMINUS, INT2FIX(1)); PUSH_INSN1(ret, location, setn, INT2FIX(4)); PUSH_SEND(ret, location, idAREF, INT2FIX(2)); - CHECK(pm_compile_pattern_match(iseq, scope_node, ((const pm_splat_node_t *) cast->rest)->expression, ret, match_failed_label, in_single_pattern, in_alternation_pattern, false, base_index + 1)); + CHECK(pm_compile_pattern_match(iseq, scope_node, ((const pm_splat_node_t *) cast->rest)->expression, ret, match_failed_label, in_single_pattern, false, base_index + 1)); } else if (posts_size > 0) { PUSH_INSN(ret, location, dup); @@ -2430,7 +2748,7 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t PUSH_INSN1(ret, location, topn, INT2FIX(3)); PUSH_SEND(ret, location, idPLUS, INT2FIX(1)); PUSH_SEND(ret, location, idAREF, INT2FIX(1)); - CHECK(pm_compile_pattern_match(iseq, scope_node, post, ret, match_failed_label, in_single_pattern, in_alternation_pattern, false, base_index + 1)); + CHECK(pm_compile_pattern_match(iseq, scope_node, post, ret, match_failed_label, in_single_pattern, false, base_index + 1)); } PUSH_INSN(ret, location, pop); @@ -2447,7 +2765,12 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t PUSH_LABEL(ret, type_error_label); PUSH_INSN1(ret, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); PUSH_INSN1(ret, location, putobject, rb_eTypeError); - PUSH_INSN1(ret, location, putobject, rb_fstring_lit("deconstruct must return Array")); + + { + VALUE operand = rb_fstring_lit("deconstruct must return Array"); + PUSH_INSN1(ret, location, putobject, operand); + } + PUSH_SEND(ret, location, id_core_raise, INT2FIX(2)); PUSH_INSN(ret, location, pop); @@ -2522,22 +2845,20 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t } PUSH_SEND(ret, location, idAREF, INT2FIX(1)); - CHECK(pm_compile_pattern_match(iseq, scope_node, cast->requireds.nodes[index], ret, next_loop_label, in_single_pattern, in_alternation_pattern, false, base_index + 4)); + CHECK(pm_compile_pattern_match(iseq, scope_node, cast->requireds.nodes[index], ret, next_loop_label, in_single_pattern, false, base_index + 4)); } - RUBY_ASSERT(PM_NODE_TYPE_P(cast->left, PM_SPLAT_NODE)); - const pm_splat_node_t *left = (const pm_splat_node_t *) cast->left; + const pm_splat_node_t *left = cast->left; if (left->expression != NULL) { PUSH_INSN1(ret, location, topn, INT2FIX(3)); PUSH_INSN1(ret, location, putobject, INT2FIX(0)); PUSH_INSN1(ret, location, topn, INT2FIX(2)); PUSH_SEND(ret, location, idAREF, INT2FIX(2)); - CHECK(pm_compile_pattern_match(iseq, scope_node, left->expression, ret, find_failed_label, in_single_pattern, in_alternation_pattern, false, base_index + 4)); + CHECK(pm_compile_pattern_match(iseq, scope_node, left->expression, ret, find_failed_label, in_single_pattern, false, base_index + 4)); } - RUBY_ASSERT(PM_NODE_TYPE_P(cast->right, PM_SPLAT_NODE)); - const pm_splat_node_t *right = (const pm_splat_node_t *) cast->right; + const pm_splat_node_t *right = cast->right; if (right->expression != NULL) { PUSH_INSN1(ret, location, topn, INT2FIX(3)); @@ -2546,7 +2867,7 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t PUSH_SEND(ret, location, idPLUS, INT2FIX(1)); PUSH_INSN1(ret, location, topn, INT2FIX(3)); PUSH_SEND(ret, location, idAREF, INT2FIX(2)); - pm_compile_pattern_match(iseq, scope_node, right->expression, ret, find_failed_label, in_single_pattern, in_alternation_pattern, false, base_index + 4); + pm_compile_pattern_match(iseq, scope_node, right->expression, ret, find_failed_label, in_single_pattern, false, base_index + 4); } PUSH_INSNL(ret, location, jump, find_succeeded_label); @@ -2560,7 +2881,12 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t PUSH_INSN1(ret, location, adjuststack, INT2FIX(3)); if (in_single_pattern) { PUSH_INSN1(ret, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); - PUSH_INSN1(ret, location, putobject, rb_fstring_lit("%p does not match to find pattern")); + + { + VALUE operand = rb_fstring_lit("%p does not match to find pattern"); + PUSH_INSN1(ret, location, putobject, operand); + } + PUSH_INSN1(ret, location, topn, INT2FIX(2)); PUSH_SEND(ret, location, id_core_sprintf, INT2FIX(2)); PUSH_INSN1(ret, location, setn, INT2FIX(base_index + PM_PATTERN_BASE_INDEX_OFFSET_ERROR_STRING + 1)); @@ -2585,7 +2911,12 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t PUSH_LABEL(ret, type_error_label); PUSH_INSN1(ret, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); PUSH_INSN1(ret, location, putobject, rb_eTypeError); - PUSH_INSN1(ret, location, putobject, rb_fstring_lit("deconstruct must return Array")); + + { + VALUE operand = rb_fstring_lit("deconstruct must return Array"); + PUSH_INSN1(ret, location, putobject, operand); + } + PUSH_SEND(ret, location, id_core_raise, INT2FIX(2)); PUSH_INSN(ret, location, pop); @@ -2634,7 +2965,12 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t } PUSH_INSN(ret, location, dup); - PUSH_INSN1(ret, location, putobject, ID2SYM(rb_intern("deconstruct_keys"))); + + { + VALUE operand = ID2SYM(rb_intern("deconstruct_keys")); + PUSH_INSN1(ret, location, putobject, operand); + } + PUSH_SEND(ret, location, idRespond_to, INT2FIX(1)); if (in_single_pattern) { CHECK(pm_compile_pattern_generic_error(iseq, scope_node, node, ret, rb_fstring_lit("%p does not respond to #deconstruct_keys"), base_index + 1)); @@ -2645,8 +2981,10 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t PUSH_INSN(ret, location, putnil); } else { + rb_obj_hide(keys); + RB_OBJ_SET_FROZEN_SHAREABLE(keys); PUSH_INSN1(ret, location, duparray, keys); - RB_OBJ_WRITTEN(iseq, Qundef, rb_obj_hide(keys)); + RB_OBJ_WRITTEN(iseq, Qundef, keys); } PUSH_SEND(ret, location, rb_intern("deconstruct_keys"), INT2FIX(1)); @@ -2660,7 +2998,6 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t if (has_keys) { DECL_ANCHOR(match_values); - INIT_ANCHOR(match_values); for (size_t index = 0; index < cast->elements.size; index++) { const pm_node_t *element = cast->elements.nodes[index]; @@ -2681,7 +3018,12 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t PUSH_INSN(ret, location, dup); PUSH_INSNL(ret, location, branchif, match_succeeded_label); - PUSH_INSN1(ret, location, putobject, rb_str_freeze(rb_sprintf("key not found: %+"PRIsVALUE, symbol))); + { + VALUE operand = rb_str_freeze(rb_sprintf("key not found: %+"PRIsVALUE, symbol)); + RB_OBJ_SET_SHAREABLE(operand); + PUSH_INSN1(ret, location, putobject, operand); + } + PUSH_INSN1(ret, location, setn, INT2FIX(base_index + PM_PATTERN_BASE_INDEX_OFFSET_ERROR_STRING + 2)); PUSH_INSN1(ret, location, putobject, Qtrue); PUSH_INSN1(ret, location, setn, INT2FIX(base_index + PM_PATTERN_BASE_INDEX_OFFSET_KEY_ERROR_P + 3)); @@ -2704,7 +3046,7 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t value = ((const pm_implicit_node_t *) value)->value; } - CHECK(pm_compile_pattern_match(iseq, scope_node, value, match_values, match_failed_label, in_single_pattern, in_alternation_pattern, false, base_index + 1)); + CHECK(pm_compile_pattern_match(iseq, scope_node, value, match_values, match_failed_label, in_single_pattern, false, base_index + 1)); } PUSH_SEQ(ret, match_values); @@ -2732,7 +3074,7 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t case PM_ASSOC_SPLAT_NODE: { const pm_assoc_splat_node_t *splat = (const pm_assoc_splat_node_t *) cast->rest; PUSH_INSN(ret, location, dup); - pm_compile_pattern_match(iseq, scope_node, splat->value, ret, match_failed_label, in_single_pattern, in_alternation_pattern, false, base_index + 1); + pm_compile_pattern_match(iseq, scope_node, splat->value, ret, match_failed_label, in_single_pattern, false, base_index + 1); break; } default: @@ -2748,7 +3090,12 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t PUSH_LABEL(ret, type_error_label); PUSH_INSN1(ret, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); PUSH_INSN1(ret, location, putobject, rb_eTypeError); - PUSH_INSN1(ret, location, putobject, rb_fstring_lit("deconstruct_keys must return Hash")); + + { + VALUE operand = rb_fstring_lit("deconstruct_keys must return Hash"); + PUSH_INSN1(ret, location, putobject, operand); + } + PUSH_SEND(ret, location, id_core_raise, INT2FIX(2)); PUSH_INSN(ret, location, pop); @@ -2772,8 +3119,8 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t LABEL *match_failed_label = NEW_LABEL(location.line); PUSH_INSN(ret, location, dup); - CHECK(pm_compile_pattern_match(iseq, scope_node, cast->value, ret, match_failed_label, in_single_pattern, in_alternation_pattern, use_deconstructed_cache, base_index + 1)); - CHECK(pm_compile_pattern(iseq, scope_node, cast->target, ret, matched_label, match_failed_label, in_single_pattern, in_alternation_pattern, false, base_index)); + CHECK(pm_compile_pattern_match(iseq, scope_node, cast->value, ret, match_failed_label, in_single_pattern, use_deconstructed_cache, base_index + 1)); + CHECK(pm_compile_pattern(iseq, scope_node, (const pm_node_t *) cast->target, ret, matched_label, match_failed_label, in_single_pattern, false, base_index)); PUSH_INSN(ret, location, putnil); PUSH_LABEL(ret, match_failed_label); @@ -2789,20 +3136,6 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t const pm_local_variable_target_node_t *cast = (const pm_local_variable_target_node_t *) node; pm_local_index_t index = pm_lookup_local_index(iseq, scope_node, cast->name, cast->depth); - // If this local variable is being written from within an alternation - // pattern, then it cannot actually be added to the local table since - // it's ambiguous which value should be used. So instead we indicate - // this with a compile error. - if (in_alternation_pattern) { - ID id = pm_constant_id_lookup(scope_node, cast->name); - const char *name = rb_id2name(id); - - if (name && strlen(name) > 0 && name[0] != '_') { - COMPILE_ERROR(iseq, location.line, "illegal variable in alternative pattern (%"PRIsVALUE")", rb_id2str(id)); - return COMPILE_NG; - } - } - PUSH_SETLOCAL(ret, location, index.index, index.level); PUSH_INSNL(ret, location, jump, matched_label); break; @@ -2818,7 +3151,7 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t // First, we're going to attempt to match against the left pattern. If // that pattern matches, then we'll skip matching the right pattern. PUSH_INSN(ret, location, dup); - CHECK(pm_compile_pattern(iseq, scope_node, cast->left, ret, matched_left_label, unmatched_left_label, in_single_pattern, true, true, base_index + 1)); + CHECK(pm_compile_pattern(iseq, scope_node, cast->left, ret, matched_left_label, unmatched_left_label, in_single_pattern, use_deconstructed_cache, base_index + 1)); // If we get here, then we matched on the left pattern. In this case we // should pop out the duplicate value that we preemptively added to @@ -2831,7 +3164,7 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t // If we get here, then we didn't match on the left pattern. In this // case we attempt to match against the right pattern. PUSH_LABEL(ret, unmatched_left_label); - CHECK(pm_compile_pattern(iseq, scope_node, cast->right, ret, matched_label, unmatched_label, in_single_pattern, true, true, base_index)); + CHECK(pm_compile_pattern(iseq, scope_node, cast->right, ret, matched_label, unmatched_label, in_single_pattern, use_deconstructed_cache, base_index)); break; } case PM_PARENTHESES_NODE: @@ -2839,7 +3172,7 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t // they do nothing since they can only wrap individual expressions and // not groups. In this case we'll recurse back into this same function // with the body of the parentheses. - return pm_compile_pattern(iseq, scope_node, ((const pm_parentheses_node_t *) node)->body, ret, matched_label, unmatched_label, in_single_pattern, in_alternation_pattern, use_deconstructed_cache, base_index); + return pm_compile_pattern(iseq, scope_node, ((const pm_parentheses_node_t *) node)->body, ret, matched_label, unmatched_label, in_single_pattern, use_deconstructed_cache, base_index); case PM_PINNED_EXPRESSION_NODE: // Pinned expressions are a way to match against the value of an // expression that should be evaluated at runtime. This looks like: @@ -2856,6 +3189,7 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t case PM_GLOBAL_VARIABLE_READ_NODE: case PM_IMAGINARY_NODE: case PM_INSTANCE_VARIABLE_READ_NODE: + case PM_IT_LOCAL_VARIABLE_READ_NODE: case PM_INTEGER_NODE: case PM_INTERPOLATED_REGULAR_EXPRESSION_NODE: case PM_INTERPOLATED_STRING_NODE: @@ -2899,7 +3233,7 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t // looks like: foo in ^@bar. To compile these, we compile the variable // that they hold. const pm_pinned_variable_node_t *cast = (const pm_pinned_variable_node_t *) node; - CHECK(pm_compile_pattern(iseq, scope_node, cast->variable, ret, matched_label, unmatched_label, in_single_pattern, in_alternation_pattern, true, base_index)); + CHECK(pm_compile_pattern(iseq, scope_node, cast->variable, ret, matched_label, unmatched_label, in_single_pattern, true, base_index)); break; } case PM_IF_NODE: @@ -2933,7 +3267,7 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t statement = cast->statements->body.nodes[0]; } - CHECK(pm_compile_pattern_match(iseq, scope_node, statement, ret, unmatched_label, in_single_pattern, in_alternation_pattern, use_deconstructed_cache, base_index)); + CHECK(pm_compile_pattern_match(iseq, scope_node, statement, ret, unmatched_label, in_single_pattern, use_deconstructed_cache, base_index)); PM_COMPILE_NOT_POPPED(predicate); if (in_single_pattern) { @@ -2947,7 +3281,11 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t PUSH_INSNL(ret, location, branchunless, match_succeeded_label); } - PUSH_INSN1(ret, location, putobject, rb_fstring_lit("guard clause does not return true")); + { + VALUE operand = rb_fstring_lit("guard clause does not return true"); + PUSH_INSN1(ret, location, putobject, operand); + } + PUSH_INSN1(ret, location, setn, INT2FIX(base_index + PM_PATTERN_BASE_INDEX_OFFSET_ERROR_STRING + 1)); PUSH_INSN1(ret, location, putobject, Qfalse); PUSH_INSN1(ret, location, setn, INT2FIX(base_index + PM_PATTERN_BASE_INDEX_OFFSET_KEY_ERROR_P + 2)); @@ -2972,7 +3310,7 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t // If we get here, then we have a node type that should not be in this // position. This would be a bug in the parser, because a different node // type should never have been created in this position in the tree. - rb_bug("Unexpected node type in pattern matching expression: %s", pm_node_type_to_str(PM_NODE_TYPE(node))); + rb_bug("Unexpected node type in pattern matching expression: %s", pm_node_type(PM_NODE_TYPE(node))); break; } @@ -2989,25 +3327,27 @@ pm_compile_pattern(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_t void pm_scope_node_init(const pm_node_t *node, pm_scope_node_t *scope, pm_scope_node_t *previous) { - // This is very important, otherwise the scope node could be seen as having - // certain flags set that _should not_ be set. - memset(scope, 0, sizeof(pm_scope_node_t)); + if (previous) { + // Copy inherited fields from the parent scope in one shot, then + // zero out the fields that are scope-specific. + *scope = *previous; + scope->locals = (pm_constant_id_list_t) { 0 }; + scope->parameters = NULL; + scope->body = NULL; + scope->local_table_for_iseq_size = 0; + scope->index_lookup_table = (pm_index_lookup_table_t) PM_INDEX_LOOKUP_TABLE_INIT; + scope->pre_execution_anchor = NULL; + } + else { + memset(scope, 0, sizeof(pm_scope_node_t)); + } scope->base.type = PM_SCOPE_NODE; scope->base.location.start = node->location.start; - scope->base.location.end = node->location.end; - + scope->base.location.length = node->location.length; scope->previous = previous; scope->ast_node = (pm_node_t *) node; - if (previous) { - scope->parser = previous->parser; - scope->encoding = previous->encoding; - scope->filepath_encoding = previous->filepath_encoding; - scope->constants = previous->constants; - scope->coverage_enabled = previous->coverage_enabled; - } - switch (PM_NODE_TYPE(node)) { case PM_BLOCK_NODE: { const pm_block_node_t *cast = (const pm_block_node_t *) node; @@ -3035,7 +3375,7 @@ pm_scope_node_init(const pm_node_t *node, pm_scope_node_t *scope, pm_scope_node_ if (cast->statements != NULL) { scope->base.location.start = cast->statements->base.location.start; - scope->base.location.end = cast->statements->base.location.end; + scope->base.location.length = cast->statements->base.location.length; } break; @@ -3055,13 +3395,6 @@ pm_scope_node_init(const pm_node_t *node, pm_scope_node_t *scope, pm_scope_node_ scope->parameters = cast->parameters; scope->body = cast->body; scope->locals = cast->locals; - - if (cast->parameters != NULL) { - scope->base.location.start = cast->parameters->location.start; - } - else { - scope->base.location.start = cast->operator_loc.end; - } break; } case PM_MODULE_NODE: { @@ -3111,8 +3444,8 @@ pm_scope_node_init(const pm_node_t *node, pm_scope_node_t *scope, pm_scope_node_ void pm_scope_node_destroy(pm_scope_node_t *scope_node) { - if (scope_node->index_lookup_table) { - st_free_table(scope_node->index_lookup_table); + if (scope_node->index_lookup_table.owned) { + xfree(scope_node->index_lookup_table.values); } } @@ -3125,7 +3458,7 @@ pm_scope_node_destroy(pm_scope_node_t *scope_node) * Normally, "send" instruction is at the last. However, qcall under branch * coverage measurement adds some instructions after the "send". * - * Note that "invokesuper" appears instead of "send". + * Note that "invokesuper", "invokesuperforward" appears instead of "send". */ static void pm_compile_retry_end_label(rb_iseq_t *iseq, LINK_ANCHOR *const ret, LABEL *retry_end_l) @@ -3181,7 +3514,7 @@ pm_iseq_builtin_function_name(const pm_scope_node_t *scope_node, const pm_node_t // Compile Primitive.attr! :leaf, ... static int -pm_compile_builtin_attr(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, const pm_arguments_node_t *arguments, const pm_node_location_t *node_location) +pm_compile_builtin_attr(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_arguments_node_t *arguments, const pm_node_location_t *node_location) { if (arguments == NULL) { COMPILE_ERROR(iseq, node_location->line, "attr!: no argument"); @@ -3191,12 +3524,12 @@ pm_compile_builtin_attr(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, cons const pm_node_t *argument; PM_NODE_LIST_FOREACH(&arguments->arguments, index, argument) { if (!PM_NODE_TYPE_P(argument, PM_SYMBOL_NODE)) { - COMPILE_ERROR(iseq, node_location->line, "non symbol argument to attr!: %s", pm_node_type_to_str(PM_NODE_TYPE(argument))); + COMPILE_ERROR(iseq, node_location->line, "non symbol argument to attr!: %s", pm_node_type(PM_NODE_TYPE(argument))); return COMPILE_NG; } VALUE symbol = pm_static_literal_value(iseq, argument, scope_node); - VALUE string = rb_sym_to_s(symbol); + VALUE string = rb_sym2str(symbol); if (strcmp(RSTRING_PTR(string), "leaf") == 0) { ISEQ_BODY(iseq)->builtin_attrs |= BUILTIN_ATTR_LEAF; @@ -3207,6 +3540,13 @@ pm_compile_builtin_attr(rb_iseq_t *iseq, const pm_scope_node_t *scope_node, cons else if (strcmp(RSTRING_PTR(string), "use_block") == 0) { iseq_set_use_block(iseq); } + else if (strcmp(RSTRING_PTR(string), "c_trace") == 0) { + // Let the iseq act like a C method in backtraces + ISEQ_BODY(iseq)->builtin_attrs |= BUILTIN_ATTR_C_TRACE; + } + else if (strcmp(RSTRING_PTR(string), "without_interrupts") == 0) { + ISEQ_BODY(iseq)->builtin_attrs |= BUILTIN_ATTR_WITHOUT_INTERRUPTS; + } else { COMPILE_ERROR(iseq, node_location->line, "unknown argument to attr!: %s", RSTRING_PTR(string)); return COMPILE_NG; @@ -3231,7 +3571,7 @@ pm_compile_builtin_arg(rb_iseq_t *iseq, LINK_ANCHOR *const ret, const pm_scope_n const pm_node_t *argument = arguments->arguments.nodes[0]; if (!PM_NODE_TYPE_P(argument, PM_SYMBOL_NODE)) { - COMPILE_ERROR(iseq, node_location->line, "non symbol argument to arg!: %s", pm_node_type_to_str(PM_NODE_TYPE(argument))); + COMPILE_ERROR(iseq, node_location->line, "non symbol argument to arg!: %s", pm_node_type(PM_NODE_TYPE(argument))); return COMPILE_NG; } @@ -3295,7 +3635,7 @@ pm_compile_builtin_mandatory_only_method(rb_iseq_t *iseq, pm_scope_node_t *scope pm_scope_node_t next_scope_node; pm_scope_node_init(&def.base, &next_scope_node, scope_node); - ISEQ_BODY(iseq)->mandatory_only_iseq = pm_iseq_new_with_opt( + const rb_iseq_t *mandatory_only_iseq = pm_iseq_build( &next_scope_node, rb_iseq_base_label(iseq), rb_iseq_path(iseq), @@ -3306,6 +3646,7 @@ pm_compile_builtin_mandatory_only_method(rb_iseq_t *iseq, pm_scope_node_t *scope ISEQ_TYPE_METHOD, ISEQ_COMPILE_DATA(iseq)->option ); + RB_OBJ_WRITE(iseq, &ISEQ_BODY(iseq)->mandatory_only_iseq, (VALUE)mandatory_only_iseq); pm_scope_node_destroy(&next_scope_node); return COMPILE_OK; @@ -3372,7 +3713,7 @@ retry:; if (cconst) { typedef VALUE(*builtin_func0)(void *, VALUE); - VALUE const_val = (*(builtin_func0)bf->func_ptr)(NULL, Qnil); + VALUE const_val = (*(builtin_func0)(uintptr_t)bf->func_ptr)(NULL, Qnil); PUSH_INSN1(ret, *node_location, putobject, const_val); return COMPILE_OK; } @@ -3380,7 +3721,6 @@ retry:; // fprintf(stderr, "func_name:%s -> %p\n", builtin_func, bf->func_ptr); DECL_ANCHOR(args_seq); - INIT_ANCHOR(args_seq); int flags = 0; struct rb_callinfo_kwarg *keywords = NULL; @@ -3411,9 +3751,10 @@ static void pm_compile_call(rb_iseq_t *iseq, const pm_call_node_t *call_node, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node, ID method_id, LABEL *start) { const pm_location_t *message_loc = &call_node->message_loc; - if (message_loc->start == NULL) message_loc = &call_node->base.location; + if (message_loc->length == 0) message_loc = &call_node->base.location; + + const pm_node_location_t location = PM_LOCATION_START_LOCATION(message_loc, call_node->base.node_id); - const pm_node_location_t location = PM_LOCATION_START_LOCATION(scope_node->parser, message_loc, call_node->base.node_id); LABEL *else_label = NEW_LABEL(location.line); LABEL *end_label = NEW_LABEL(location.line); LABEL *retry_end_l = NEW_LABEL(location.line); @@ -3424,18 +3765,37 @@ pm_compile_call(rb_iseq_t *iseq, const pm_call_node_t *call_node, LINK_ANCHOR *c if (PM_NODE_FLAG_P(call_node, PM_CALL_NODE_FLAGS_SAFE_NAVIGATION)) { if (PM_BRANCH_COVERAGE_P(iseq)) { - const uint8_t *cursors[3] = { - call_node->closing_loc.end, - call_node->arguments == NULL ? NULL : call_node->arguments->base.location.end, - call_node->message_loc.end - }; + uint32_t end_cursor = 0; + bool end_found = false; - const uint8_t *end_cursor = cursors[0]; - end_cursor = (end_cursor == NULL || cursors[1] == NULL) ? cursors[1] : (end_cursor > cursors[1] ? end_cursor : cursors[1]); - end_cursor = (end_cursor == NULL || cursors[2] == NULL) ? cursors[2] : (end_cursor > cursors[2] ? end_cursor : cursors[2]); + if (call_node->closing_loc.length > 0) { + uint32_t cursor = call_node->closing_loc.start + call_node->closing_loc.length; + end_cursor = cursor; + end_found = true; + } - const pm_line_column_t start_location = PM_NODE_START_LINE_COLUMN(scope_node->parser, call_node); - const pm_line_column_t end_location = pm_newline_list_line_column(&scope_node->parser->newline_list, end_cursor, scope_node->parser->start_line); + if (call_node->arguments != NULL) { + uint32_t cursor = call_node->arguments->base.location.start + call_node->arguments->base.location.length; + if (!end_found || cursor > end_cursor) { + end_cursor = cursor; + end_found = true; + } + } + + if (call_node->message_loc.length > 0) { + uint32_t cursor = call_node->message_loc.start + call_node->message_loc.length; + if (!end_found || cursor > end_cursor) { + end_cursor = cursor; + end_found = true; + } + } + + if (!end_found) { + end_cursor = call_node->closing_loc.start + call_node->closing_loc.length; + } + + const pm_line_column_t start_location = PM_NODE_START_LINE_COLUMN(call_node); + const pm_line_column_t end_location = pm_line_offset_list_line_column_cached(scope_node->line_offsets, end_cursor, scope_node->start_line, &scope_node->last_line); code_location = (rb_code_location_t) { .beg_pos = { .lineno = start_location.line, .column = start_location.column }, @@ -3451,6 +3811,8 @@ pm_compile_call(rb_iseq_t *iseq, const pm_call_node_t *call_node, LINK_ANCHOR *c add_trace_branch_coverage(iseq, ret, &code_location, node_id, 0, "then", branches); } + LINK_ELEMENT *opt_new_prelude = LAST_ELEMENT(ret); + int flags = 0; struct rb_callinfo_kwarg *kw_arg = NULL; @@ -3463,7 +3825,7 @@ pm_compile_call(rb_iseq_t *iseq, const pm_call_node_t *call_node, LINK_ANCHOR *c pm_scope_node_t next_scope_node; pm_scope_node_init(call_node->block, &next_scope_node, scope_node); - block_iseq = NEW_CHILD_ISEQ(&next_scope_node, make_name_for_block(iseq), ISEQ_TYPE_BLOCK, pm_node_line_number(scope_node->parser, call_node->block)); + block_iseq = NEW_CHILD_ISEQ(&next_scope_node, make_name_for_block(iseq), ISEQ_TYPE_BLOCK, pm_node_line_number_cached(call_node->block, scope_node)); pm_scope_node_destroy(&next_scope_node); ISEQ_COMPILE_DATA(iseq)->current_block = block_iseq; } @@ -3507,7 +3869,50 @@ pm_compile_call(rb_iseq_t *iseq, const pm_call_node_t *call_node, LINK_ANCHOR *c PUSH_INSN(ret, location, splatkw); } - PUSH_SEND_R(ret, location, method_id, INT2FIX(orig_argc), block_iseq, INT2FIX(flags), kw_arg); + LABEL *not_basic_new = NEW_LABEL(location.line); + LABEL *not_basic_new_finish = NEW_LABEL(location.line); + + bool inline_new = ISEQ_COMPILE_DATA(iseq)->option->specialized_instruction && + method_id == rb_intern("new") && + call_node->block == NULL && + (flags & VM_CALL_ARGS_BLOCKARG) == 0; + + if (inline_new) { + if (LAST_ELEMENT(ret) == opt_new_prelude) { + PUSH_INSN(ret, location, putnil); + PUSH_INSN(ret, location, swap); + } + else { + ELEM_INSERT_NEXT(opt_new_prelude, &new_insn_body(iseq, location.line, location.node_id, BIN(swap), 0)->link); + ELEM_INSERT_NEXT(opt_new_prelude, &new_insn_body(iseq, location.line, location.node_id, BIN(putnil), 0)->link); + } + + // Jump unless the receiver uses the "basic" implementation of "new" + VALUE ci; + if (flags & VM_CALL_FORWARDING) { + ci = (VALUE)new_callinfo(iseq, method_id, orig_argc + 1, flags, kw_arg, 0); + } + else { + ci = (VALUE)new_callinfo(iseq, method_id, orig_argc, flags, kw_arg, 0); + } + + PUSH_INSN2(ret, location, opt_new, ci, not_basic_new); + LABEL_REF(not_basic_new); + // optimized path + PUSH_SEND_R(ret, location, rb_intern("initialize"), INT2FIX(orig_argc), block_iseq, INT2FIX(flags | VM_CALL_FCALL), kw_arg); + PUSH_INSNL(ret, location, jump, not_basic_new_finish); + + PUSH_LABEL(ret, not_basic_new); + // Fall back to normal send + PUSH_SEND_R(ret, location, method_id, INT2FIX(orig_argc), block_iseq, INT2FIX(flags), kw_arg); + PUSH_INSN(ret, location, swap); + + PUSH_LABEL(ret, not_basic_new_finish); + PUSH_INSN(ret, location, pop); + } + else { + PUSH_SEND_R(ret, location, method_id, INT2FIX(orig_argc), block_iseq, INT2FIX(flags), kw_arg); + } if (block_iseq && ISEQ_BODY(block_iseq)->catch_table) { pm_compile_retry_end_label(iseq, ret, retry_end_l); @@ -3529,264 +3934,585 @@ pm_compile_call(rb_iseq_t *iseq, const pm_call_node_t *call_node, LINK_ANCHOR *c ISEQ_COMPILE_DATA(iseq)->current_block = previous_block; } +/** + * Compile and return the VALUE associated with the given back reference read + * node. + */ +static inline VALUE +pm_compile_back_reference_ref(const pm_scope_node_t *scope_node, const pm_back_reference_read_node_t *node) +{ + const char *type = (const char *) (pm_parser_start(scope_node->parser) + node->base.location.start + 1); + + // Since a back reference is `$<char>`, Ruby represents the ID as an + // rb_intern on the value after the `$`. + return INT2FIX(rb_intern2(type, 1)) << 1 | 1; +} + +/** + * Compile and return the VALUE associated with the given numbered reference + * read node. + */ +static inline VALUE +pm_compile_numbered_reference_ref(const pm_numbered_reference_read_node_t *node) +{ + return INT2FIX(node->number << 1); +} + static void pm_compile_defined_expr0(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_location_t *node_location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node, bool in_condition, LABEL **lfinish, bool explicit_receiver) { +#define PUSH_VAL(type) (in_condition ? Qtrue : rb_iseq_defined_string(type)) + // in_condition is the same as compile.c's needstr enum defined_type dtype = DEFINED_NOT_DEFINED; const pm_node_location_t location = *node_location; switch (PM_NODE_TYPE(node)) { - case PM_ARGUMENTS_NODE: { - const pm_arguments_node_t *cast = (const pm_arguments_node_t *) node; - const pm_node_list_t *arguments = &cast->arguments; - for (size_t idx = 0; idx < arguments->size; idx++) { - const pm_node_t *argument = arguments->nodes[idx]; - pm_compile_defined_expr0(iseq, argument, node_location, ret, popped, scope_node, in_condition, lfinish, explicit_receiver); - - if (!lfinish[1]) { - lfinish[1] = NEW_LABEL(location.line); - } - PUSH_INSNL(ret, location, branchunless, lfinish[1]); - } - dtype = DEFINED_TRUE; - break; - } +/* DEFINED_NIL ****************************************************************/ case PM_NIL_NODE: + // defined?(nil) + // ^^^ dtype = DEFINED_NIL; break; - case PM_PARENTHESES_NODE: { - const pm_parentheses_node_t *cast = (const pm_parentheses_node_t *) node; +/* DEFINED_IVAR ***************************************************************/ + case PM_INSTANCE_VARIABLE_READ_NODE: { + // defined?(@a) + // ^^ + const pm_instance_variable_read_node_t *cast = (const pm_instance_variable_read_node_t *) node; + ID name = pm_constant_id_lookup(scope_node, cast->name); - if (cast->body == NULL) { - // If we have empty parentheses, then we want to return "nil". - dtype = DEFINED_NIL; - } - else if (PM_NODE_TYPE_P(cast->body, PM_STATEMENTS_NODE) && ((const pm_statements_node_t *) cast->body)->body.size == 1) { - // If we have a parentheses node that is wrapping a single statement - // then we want to recurse down to that statement and compile it. - pm_compile_defined_expr0(iseq, ((const pm_statements_node_t *) cast->body)->body.nodes[0], node_location, ret, popped, scope_node, in_condition, lfinish, explicit_receiver); - return; - } - else { - // Otherwise, we have parentheses wrapping multiple statements, in - // which case this is defined as "expression". - dtype = DEFINED_EXPR; - } + PUSH_INSN3(ret, location, definedivar, ID2SYM(name), get_ivar_ic_value(iseq, name), PUSH_VAL(DEFINED_IVAR)); - break; + return; } - case PM_SELF_NODE: - dtype = DEFINED_SELF; - break; - case PM_TRUE_NODE: - dtype = DEFINED_TRUE; - break; - case PM_FALSE_NODE: - dtype = DEFINED_FALSE; +/* DEFINED_LVAR ***************************************************************/ + case PM_LOCAL_VARIABLE_READ_NODE: + // a = 1; defined?(a) + // ^ + case PM_IT_LOCAL_VARIABLE_READ_NODE: + // 1.then { defined?(it) } + // ^^ + dtype = DEFINED_LVAR; break; - case PM_ARRAY_NODE: { - const pm_array_node_t *cast = (const pm_array_node_t *) node; - - if (cast->elements.size > 0 && !lfinish[1]) { - lfinish[1] = NEW_LABEL(location.line); - } +/* DEFINED_GVAR ***************************************************************/ + case PM_GLOBAL_VARIABLE_READ_NODE: { + // defined?($a) + // ^^ + const pm_global_variable_read_node_t *cast = (const pm_global_variable_read_node_t *) node; + ID name = pm_constant_id_lookup(scope_node, cast->name); - for (size_t index = 0; index < cast->elements.size; index++) { - pm_compile_defined_expr0(iseq, cast->elements.nodes[index], node_location, ret, popped, scope_node, true, lfinish, false); - PUSH_INSNL(ret, location, branchunless, lfinish[1]); - } + PUSH_INSN(ret, location, putnil); + PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_GVAR), ID2SYM(name), PUSH_VAL(DEFINED_GVAR)); - dtype = DEFINED_EXPR; - break; + return; } - case PM_HASH_NODE: - case PM_KEYWORD_HASH_NODE: { - const pm_node_list_t *elements; - - if (PM_NODE_TYPE_P(node, PM_HASH_NODE)) { - elements = &((const pm_hash_node_t *) node)->elements; - } - else { - elements = &((const pm_keyword_hash_node_t *) node)->elements; - } - - if (elements->size > 0 && !lfinish[1]) { - lfinish[1] = NEW_LABEL(location.line); - } +/* DEFINED_CVAR ***************************************************************/ + case PM_CLASS_VARIABLE_READ_NODE: { + // defined?(@@a) + // ^^^ + const pm_class_variable_read_node_t *cast = (const pm_class_variable_read_node_t *) node; + ID name = pm_constant_id_lookup(scope_node, cast->name); - for (size_t index = 0; index < elements->size; index++) { - const pm_node_t *element = elements->nodes[index]; + PUSH_INSN(ret, location, putnil); + PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_CVAR), ID2SYM(name), PUSH_VAL(DEFINED_CVAR)); - switch (PM_NODE_TYPE(element)) { - case PM_ASSOC_NODE: { - const pm_assoc_node_t *assoc = (const pm_assoc_node_t *) element; + return; + } +/* DEFINED_CONST **************************************************************/ + case PM_CONSTANT_READ_NODE: { + // defined?(A) + // ^ + const pm_constant_read_node_t *cast = (const pm_constant_read_node_t *) node; + ID name = pm_constant_id_lookup(scope_node, cast->name); - pm_compile_defined_expr0(iseq, assoc->key, node_location, ret, popped, scope_node, true, lfinish, false); - PUSH_INSNL(ret, location, branchunless, lfinish[1]); + PUSH_INSN(ret, location, putnil); + PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_CONST), ID2SYM(name), PUSH_VAL(DEFINED_CONST)); - pm_compile_defined_expr0(iseq, assoc->value, node_location, ret, popped, scope_node, true, lfinish, false); - PUSH_INSNL(ret, location, branchunless, lfinish[1]); + return; + } +/* DEFINED_YIELD **************************************************************/ + case PM_YIELD_NODE: + // defined?(yield) + // ^^^^^ + iseq_set_use_block(ISEQ_BODY(iseq)->local_iseq); - break; - } - case PM_ASSOC_SPLAT_NODE: { - const pm_assoc_splat_node_t *assoc_splat = (const pm_assoc_splat_node_t *) element; + PUSH_INSN(ret, location, putnil); + PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_YIELD), 0, PUSH_VAL(DEFINED_YIELD)); - pm_compile_defined_expr0(iseq, assoc_splat->value, node_location, ret, popped, scope_node, true, lfinish, false); - PUSH_INSNL(ret, location, branchunless, lfinish[1]); + return; +/* DEFINED_ZSUPER *************************************************************/ + case PM_SUPER_NODE: { + // defined?(super 1, 2) + // ^^^^^^^^^^ + const pm_super_node_t *cast = (const pm_super_node_t *) node; - break; - } - default: - rb_bug("unexpected node type in hash node: %s", pm_node_type_to_str(PM_NODE_TYPE(element))); - break; - } + if (cast->block != NULL && !PM_NODE_TYPE_P(cast->block, PM_BLOCK_ARGUMENT_NODE)) { + dtype = DEFINED_EXPR; + break; } - dtype = DEFINED_EXPR; - break; + PUSH_INSN(ret, location, putnil); + PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_ZSUPER), 0, PUSH_VAL(DEFINED_ZSUPER)); + return; } - case PM_SPLAT_NODE: { - const pm_splat_node_t *cast = (const pm_splat_node_t *) node; - pm_compile_defined_expr0(iseq, cast->expression, node_location, ret, popped, scope_node, in_condition, lfinish, false); + case PM_FORWARDING_SUPER_NODE: { + // defined?(super) + // ^^^^^ + const pm_forwarding_super_node_t *cast = (const pm_forwarding_super_node_t *) node; - if (!lfinish[1]) { - lfinish[1] = NEW_LABEL(location.line); + if (cast->block != NULL) { + dtype = DEFINED_EXPR; + break; } - PUSH_INSNL(ret, location, branchunless, lfinish[1]); - dtype = DEFINED_EXPR; - break; - } - case PM_IMPLICIT_NODE: { - const pm_implicit_node_t *cast = (const pm_implicit_node_t *) node; - pm_compile_defined_expr0(iseq, cast->value, node_location, ret, popped, scope_node, in_condition, lfinish, explicit_receiver); + PUSH_INSN(ret, location, putnil); + PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_ZSUPER), 0, PUSH_VAL(DEFINED_ZSUPER)); return; } +/* DEFINED_SELF ***************************************************************/ + case PM_SELF_NODE: + // defined?(self) + // ^^^^ + dtype = DEFINED_SELF; + break; +/* DEFINED_TRUE ***************************************************************/ + case PM_TRUE_NODE: + // defined?(true) + // ^^^^ + dtype = DEFINED_TRUE; + break; +/* DEFINED_FALSE **************************************************************/ + case PM_FALSE_NODE: + // defined?(false) + // ^^^^^ + dtype = DEFINED_FALSE; + break; +/* DEFINED_ASGN ***************************************************************/ + case PM_CALL_AND_WRITE_NODE: + // defined?(a.a &&= 1) + // ^^^^^^^^^ + case PM_CALL_OPERATOR_WRITE_NODE: + // defined?(a.a += 1) + // ^^^^^^^^ + case PM_CALL_OR_WRITE_NODE: + // defined?(a.a ||= 1) + // ^^^^^^^^^ + case PM_CLASS_VARIABLE_AND_WRITE_NODE: + // defined?(@@a &&= 1) + // ^^^^^^^^^ + case PM_CLASS_VARIABLE_OPERATOR_WRITE_NODE: + // defined?(@@a += 1) + // ^^^^^^^^ + case PM_CLASS_VARIABLE_OR_WRITE_NODE: + // defined?(@@a ||= 1) + // ^^^^^^^^^ + case PM_CLASS_VARIABLE_WRITE_NODE: + // defined?(@@a = 1) + // ^^^^^^^ + case PM_CONSTANT_AND_WRITE_NODE: + // defined?(A &&= 1) + // ^^^^^^^ + case PM_CONSTANT_OPERATOR_WRITE_NODE: + // defined?(A += 1) + // ^^^^^^ + case PM_CONSTANT_OR_WRITE_NODE: + // defined?(A ||= 1) + // ^^^^^^^ + case PM_CONSTANT_PATH_AND_WRITE_NODE: + // defined?(A::A &&= 1) + // ^^^^^^^^^^ + case PM_CONSTANT_PATH_OPERATOR_WRITE_NODE: + // defined?(A::A += 1) + // ^^^^^^^^^ + case PM_CONSTANT_PATH_OR_WRITE_NODE: + // defined?(A::A ||= 1) + // ^^^^^^^^^^ + case PM_CONSTANT_PATH_WRITE_NODE: + // defined?(A::A = 1) + // ^^^^^^^^ + case PM_CONSTANT_WRITE_NODE: + // defined?(A = 1) + // ^^^^^ + case PM_GLOBAL_VARIABLE_AND_WRITE_NODE: + // defined?($a &&= 1) + // ^^^^^^^^ + case PM_GLOBAL_VARIABLE_OPERATOR_WRITE_NODE: + // defined?($a += 1) + // ^^^^^^^ + case PM_GLOBAL_VARIABLE_OR_WRITE_NODE: + // defined?($a ||= 1) + // ^^^^^^^^ + case PM_GLOBAL_VARIABLE_WRITE_NODE: + // defined?($a = 1) + // ^^^^^^ + case PM_INDEX_AND_WRITE_NODE: + // defined?(a[1] &&= 1) + // ^^^^^^^^^^ + case PM_INDEX_OPERATOR_WRITE_NODE: + // defined?(a[1] += 1) + // ^^^^^^^^^ + case PM_INDEX_OR_WRITE_NODE: + // defined?(a[1] ||= 1) + // ^^^^^^^^^^ + case PM_INSTANCE_VARIABLE_AND_WRITE_NODE: + // defined?(@a &&= 1) + // ^^^^^^^^ + case PM_INSTANCE_VARIABLE_OPERATOR_WRITE_NODE: + // defined?(@a += 1) + // ^^^^^^^ + case PM_INSTANCE_VARIABLE_OR_WRITE_NODE: + // defined?(@a ||= 1) + // ^^^^^^^^ + case PM_INSTANCE_VARIABLE_WRITE_NODE: + // defined?(@a = 1) + // ^^^^^^ + case PM_LOCAL_VARIABLE_AND_WRITE_NODE: + // defined?(a &&= 1) + // ^^^^^^^ + case PM_LOCAL_VARIABLE_OPERATOR_WRITE_NODE: + // defined?(a += 1) + // ^^^^^^ + case PM_LOCAL_VARIABLE_OR_WRITE_NODE: + // defined?(a ||= 1) + // ^^^^^^^ + case PM_LOCAL_VARIABLE_WRITE_NODE: + // defined?(a = 1) + // ^^^^^ + case PM_MULTI_WRITE_NODE: + // defined?((a, = 1)) + // ^^^^^^ + dtype = DEFINED_ASGN; + break; +/* DEFINED_EXPR ***************************************************************/ + case PM_ALIAS_GLOBAL_VARIABLE_NODE: + // defined?((alias $a $b)) + // ^^^^^^^^^^^ + case PM_ALIAS_METHOD_NODE: + // defined?((alias a b)) + // ^^^^^^^^^ case PM_AND_NODE: - case PM_BEGIN_NODE: + // defined?(a and b) + // ^^^^^^^ case PM_BREAK_NODE: - case PM_CASE_NODE: + // defined?(break 1) + // ^^^^^^^ case PM_CASE_MATCH_NODE: + // defined?(case 1; in 1; end) + // ^^^^^^^^^^^^^^^^^ + case PM_CASE_NODE: + // defined?(case 1; when 1; end) + // ^^^^^^^^^^^^^^^^^^^ case PM_CLASS_NODE: + // defined?(class Foo; end) + // ^^^^^^^^^^^^^^ case PM_DEF_NODE: + // defined?(def a() end) + // ^^^^^^^^^^^ case PM_DEFINED_NODE: + // defined?(defined?(a)) + // ^^^^^^^^^^^ + case PM_FLIP_FLOP_NODE: + // defined?(not (a .. b)) + // ^^^^^^ case PM_FLOAT_NODE: + // defined?(1.0) + // ^^^ case PM_FOR_NODE: + // defined?(for a in 1 do end) + // ^^^^^^^^^^^^^^^^^ case PM_IF_NODE: + // defined?(if a then end) + // ^^^^^^^^^^^^^ case PM_IMAGINARY_NODE: + // defined?(1i) + // ^^ case PM_INTEGER_NODE: + // defined?(1) + // ^ + case PM_INTERPOLATED_MATCH_LAST_LINE_NODE: + // defined?(not /#{1}/) + // ^^^^^^ case PM_INTERPOLATED_REGULAR_EXPRESSION_NODE: + // defined?(/#{1}/) + // ^^^^^^ case PM_INTERPOLATED_STRING_NODE: + // defined?("#{1}") + // ^^^^^^ case PM_INTERPOLATED_SYMBOL_NODE: + // defined?(:"#{1}") + // ^^^^^^^ case PM_INTERPOLATED_X_STRING_NODE: + // defined?(`#{1}`) + // ^^^^^^ case PM_LAMBDA_NODE: + // defined?(-> {}) + // ^^^^^ + case PM_MATCH_LAST_LINE_NODE: + // defined?(not //) + // ^^^^^^ case PM_MATCH_PREDICATE_NODE: + // defined?(1 in 1) + // ^^^^^^ case PM_MATCH_REQUIRED_NODE: + // defined?(1 => 1) + // ^^^^^^ case PM_MATCH_WRITE_NODE: + // defined?(/(?<a>)/ =~ "") + // ^^^^^^^^^^^^^^ case PM_MODULE_NODE: + // defined?(module A end) + // ^^^^^^^^^^^^ case PM_NEXT_NODE: + // defined?(next 1) + // ^^^^^^ case PM_OR_NODE: + // defined?(a or b) + // ^^^^^^ + case PM_POST_EXECUTION_NODE: + // defined?((END {})) + // ^^^^^^^^ case PM_RANGE_NODE: + // defined?(1..1) + // ^^^^ case PM_RATIONAL_NODE: + // defined?(1r) + // ^^ case PM_REDO_NODE: + // defined?(redo) + // ^^^^ case PM_REGULAR_EXPRESSION_NODE: + // defined?(//) + // ^^ + case PM_RESCUE_MODIFIER_NODE: + // defined?(a rescue b) + // ^^^^^^^^^^ case PM_RETRY_NODE: + // defined?(retry) + // ^^^^^ case PM_RETURN_NODE: + // defined?(return) + // ^^^^^^ case PM_SINGLETON_CLASS_NODE: + // defined?(class << self; end) + // ^^^^^^^^^^^^^^^^^^ case PM_SOURCE_ENCODING_NODE: + // defined?(__ENCODING__) + // ^^^^^^^^^^^^ case PM_SOURCE_FILE_NODE: + // defined?(__FILE__) + // ^^^^^^^^ case PM_SOURCE_LINE_NODE: + // defined?(__LINE__) + // ^^^^^^^^ case PM_STRING_NODE: + // defined?("") + // ^^ case PM_SYMBOL_NODE: + // defined?(:a) + // ^^ + case PM_UNDEF_NODE: + // defined?((undef a)) + // ^^^^^^^ case PM_UNLESS_NODE: + // defined?(unless a then end) + // ^^^^^^^^^^^^^^^^^ case PM_UNTIL_NODE: + // defined?(until a do end) + // ^^^^^^^^^^^^^^ case PM_WHILE_NODE: + // defined?(while a do end) + // ^^^^^^^^^^^^^^ case PM_X_STRING_NODE: + // defined?(``) + // ^^ dtype = DEFINED_EXPR; break; - case PM_LOCAL_VARIABLE_READ_NODE: - dtype = DEFINED_LVAR; - break; - -#define PUSH_VAL(type) (in_condition ? Qtrue : rb_iseq_defined_string(type)) - - case PM_INSTANCE_VARIABLE_READ_NODE: { - const pm_instance_variable_read_node_t *cast = (const pm_instance_variable_read_node_t *) node; - - ID name = pm_constant_id_lookup(scope_node, cast->name); - PUSH_INSN3(ret, location, definedivar, ID2SYM(name), get_ivar_ic_value(iseq, name), PUSH_VAL(DEFINED_IVAR)); - - return; - } +/* DEFINED_REF ****************************************************************/ case PM_BACK_REFERENCE_READ_NODE: { - const char *char_ptr = (const char *) (node->location.start + 1); - ID backref_val = INT2FIX(rb_intern2(char_ptr, 1)) << 1 | 1; + // defined?($+) + // ^^ + const pm_back_reference_read_node_t *cast = (const pm_back_reference_read_node_t *) node; + VALUE ref = pm_compile_back_reference_ref(scope_node, cast); PUSH_INSN(ret, location, putnil); - PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_REF), backref_val, PUSH_VAL(DEFINED_GVAR)); + PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_REF), ref, PUSH_VAL(DEFINED_GVAR)); return; } case PM_NUMBERED_REFERENCE_READ_NODE: { - uint32_t reference_number = ((const pm_numbered_reference_read_node_t *) node)->number; + // defined?($1) + // ^^ + const pm_numbered_reference_read_node_t *cast = (const pm_numbered_reference_read_node_t *) node; + VALUE ref = pm_compile_numbered_reference_ref(cast); PUSH_INSN(ret, location, putnil); - PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_REF), INT2FIX(reference_number << 1), PUSH_VAL(DEFINED_GVAR)); + PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_REF), ref, PUSH_VAL(DEFINED_GVAR)); return; } - case PM_GLOBAL_VARIABLE_READ_NODE: { - const pm_global_variable_read_node_t *cast = (const pm_global_variable_read_node_t *) node; - VALUE name = ID2SYM(pm_constant_id_lookup(scope_node, cast->name)); +/* DEFINED_CONST_FROM *********************************************************/ + case PM_CONSTANT_PATH_NODE: { + // defined?(A::A) + // ^^^^ + const pm_constant_path_node_t *cast = (const pm_constant_path_node_t *) node; + ID name = pm_constant_id_lookup(scope_node, cast->name); - PUSH_INSN(ret, location, putnil); - PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_GVAR), name, PUSH_VAL(DEFINED_GVAR)); + if (cast->parent != NULL) { + if (!lfinish[1]) lfinish[1] = NEW_LABEL(location.line); + pm_compile_defined_expr0(iseq, cast->parent, node_location, ret, popped, scope_node, true, lfinish, false); + + PUSH_INSNL(ret, location, branchunless, lfinish[1]); + PM_COMPILE(cast->parent); + } + else { + PUSH_INSN1(ret, location, putobject, rb_cObject); + } + PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_CONST_FROM), ID2SYM(name), PUSH_VAL(DEFINED_CONST)); return; } - case PM_CLASS_VARIABLE_READ_NODE: { - const pm_class_variable_read_node_t *cast = (const pm_class_variable_read_node_t *) node; - VALUE name = ID2SYM(pm_constant_id_lookup(scope_node, cast->name)); +/* Containers *****************************************************************/ + case PM_BEGIN_NODE: { + // defined?(begin end) + // ^^^^^^^^^ + const pm_begin_node_t *cast = (const pm_begin_node_t *) node; - PUSH_INSN(ret, location, putnil); - PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_CVAR), name, PUSH_VAL(DEFINED_CVAR)); + if (cast->rescue_clause == NULL && cast->ensure_clause == NULL && cast->else_clause == NULL) { + if (cast->statements == NULL) { + // If we have empty statements, then we want to return "nil". + dtype = DEFINED_NIL; + } + else if (cast->statements->body.size == 1) { + // If we have a begin node that is wrapping a single statement + // then we want to recurse down to that statement and compile + // it. + pm_compile_defined_expr0(iseq, cast->statements->body.nodes[0], node_location, ret, popped, scope_node, in_condition, lfinish, false); + return; + } + else { + // Otherwise, we have a begin wrapping multiple statements, in + // which case this is defined as "expression". + dtype = DEFINED_EXPR; + } + } else { + // If we have any of the other clauses besides the main begin/end, + // this is defined as "expression". + dtype = DEFINED_EXPR; + } - return; + break; } - case PM_CONSTANT_READ_NODE: { - const pm_constant_read_node_t *cast = (const pm_constant_read_node_t *) node; - VALUE name = ID2SYM(pm_constant_id_lookup(scope_node, cast->name)); + case PM_PARENTHESES_NODE: { + // defined?(()) + // ^^ + const pm_parentheses_node_t *cast = (const pm_parentheses_node_t *) node; - PUSH_INSN(ret, location, putnil); - PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_CONST), name, PUSH_VAL(DEFINED_CONST)); + if (cast->body == NULL) { + // If we have empty parentheses, then we want to return "nil". + dtype = DEFINED_NIL; + } + else if (PM_NODE_TYPE_P(cast->body, PM_STATEMENTS_NODE) && !PM_NODE_FLAG_P(cast, PM_PARENTHESES_NODE_FLAGS_MULTIPLE_STATEMENTS)) { + // If we have a parentheses node that is wrapping a single statement + // then we want to recurse down to that statement and compile it. + pm_compile_defined_expr0(iseq, ((const pm_statements_node_t *) cast->body)->body.nodes[0], node_location, ret, popped, scope_node, in_condition, lfinish, false); + return; + } + else { + // Otherwise, we have parentheses wrapping multiple statements, in + // which case this is defined as "expression". + dtype = DEFINED_EXPR; + } - return; + break; } - case PM_CONSTANT_PATH_NODE: { - const pm_constant_path_node_t *cast = (const pm_constant_path_node_t *) node; - VALUE name = ID2SYM(pm_constant_id_lookup(scope_node, cast->name)); + case PM_ARRAY_NODE: { + // defined?([]) + // ^^ + const pm_array_node_t *cast = (const pm_array_node_t *) node; - if (cast->parent != NULL) { - if (!lfinish[1]) lfinish[1] = NEW_LABEL(location.line); - pm_compile_defined_expr0(iseq, cast->parent, node_location, ret, popped, scope_node, true, lfinish, false); + if (cast->elements.size > 0 && !lfinish[1]) { + lfinish[1] = NEW_LABEL(location.line); + } + for (size_t index = 0; index < cast->elements.size; index++) { + pm_compile_defined_expr0(iseq, cast->elements.nodes[index], node_location, ret, popped, scope_node, true, lfinish, false); PUSH_INSNL(ret, location, branchunless, lfinish[1]); - PM_COMPILE(cast->parent); + } + + dtype = DEFINED_EXPR; + break; + } + case PM_HASH_NODE: + // defined?({ a: 1 }) + // ^^^^^^^^ + case PM_KEYWORD_HASH_NODE: { + // defined?(a(a: 1)) + // ^^^^ + const pm_node_list_t *elements; + + if (PM_NODE_TYPE_P(node, PM_HASH_NODE)) { + elements = &((const pm_hash_node_t *) node)->elements; } else { - PUSH_INSN1(ret, location, putobject, rb_cObject); + elements = &((const pm_keyword_hash_node_t *) node)->elements; + } + + if (elements->size > 0 && !lfinish[1]) { + lfinish[1] = NEW_LABEL(location.line); + } + + for (size_t index = 0; index < elements->size; index++) { + pm_compile_defined_expr0(iseq, elements->nodes[index], node_location, ret, popped, scope_node, true, lfinish, false); + PUSH_INSNL(ret, location, branchunless, lfinish[1]); + } + + dtype = DEFINED_EXPR; + break; + } + case PM_ASSOC_NODE: { + // defined?({ a: 1 }) + // ^^^^ + const pm_assoc_node_t *cast = (const pm_assoc_node_t *) node; + + pm_compile_defined_expr0(iseq, cast->key, node_location, ret, popped, scope_node, true, lfinish, false); + PUSH_INSNL(ret, location, branchunless, lfinish[1]); + pm_compile_defined_expr0(iseq, cast->value, node_location, ret, popped, scope_node, true, lfinish, false); + + return; + } + case PM_ASSOC_SPLAT_NODE: { + // defined?({ **a }) + // ^^^^ + const pm_assoc_splat_node_t *cast = (const pm_assoc_splat_node_t *) node; + + if (cast->value == NULL) { + dtype = DEFINED_EXPR; + break; } - PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_CONST_FROM), name, PUSH_VAL(DEFINED_CONST)); + pm_compile_defined_expr0(iseq, cast->value, node_location, ret, popped, scope_node, true, lfinish, false); + return; + } + case PM_IMPLICIT_NODE: { + // defined?({ a: }) + // ^^ + const pm_implicit_node_t *cast = (const pm_implicit_node_t *) node; + pm_compile_defined_expr0(iseq, cast->value, node_location, ret, popped, scope_node, in_condition, lfinish, false); return; } case PM_CALL_NODE: { +#define BLOCK_P(cast) ((cast)->block != NULL && PM_NODE_TYPE_P((cast)->block, PM_BLOCK_NODE)) + + // defined?(a(1, 2, 3)) + // ^^^^^^^^^^ const pm_call_node_t *cast = ((const pm_call_node_t *) node); - ID method_id = pm_constant_id_lookup(scope_node, cast->name); - if (cast->receiver || cast->arguments) { + if (BLOCK_P(cast)) { + dtype = DEFINED_EXPR; + break; + } + + if (cast->receiver || cast->arguments || (cast->block && PM_NODE_TYPE_P(cast->block, PM_BLOCK_ARGUMENT_NODE))) { if (!lfinish[1]) lfinish[1] = NEW_LABEL(location.line); if (!lfinish[2]) lfinish[2] = NEW_LABEL(location.line); } @@ -3796,93 +4522,155 @@ pm_compile_defined_expr0(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_l PUSH_INSNL(ret, location, branchunless, lfinish[1]); } - if (cast->receiver) { - pm_compile_defined_expr0(iseq, cast->receiver, node_location, ret, popped, scope_node, true, lfinish, true); + if (cast->block && PM_NODE_TYPE_P(cast->block, PM_BLOCK_ARGUMENT_NODE)) { + pm_compile_defined_expr0(iseq, cast->block, node_location, ret, popped, scope_node, true, lfinish, false); + PUSH_INSNL(ret, location, branchunless, lfinish[1]); + } - if (PM_NODE_TYPE_P(cast->receiver, PM_CALL_NODE)) { + if (cast->receiver) { + if (PM_NODE_TYPE_P(cast->receiver, PM_CALL_NODE) && !BLOCK_P((const pm_call_node_t *) cast->receiver)) { + // Special behavior here where we chain calls together. This is + // the only path that sets explicit_receiver to true. + pm_compile_defined_expr0(iseq, cast->receiver, node_location, ret, popped, scope_node, true, lfinish, true); PUSH_INSNL(ret, location, branchunless, lfinish[2]); const pm_call_node_t *receiver = (const pm_call_node_t *) cast->receiver; ID method_id = pm_constant_id_lookup(scope_node, receiver->name); + pm_compile_call(iseq, receiver, ret, popped, scope_node, method_id, NULL); } else { + pm_compile_defined_expr0(iseq, cast->receiver, node_location, ret, popped, scope_node, true, lfinish, false); PUSH_INSNL(ret, location, branchunless, lfinish[1]); PM_COMPILE(cast->receiver); } + ID method_id = pm_constant_id_lookup(scope_node, cast->name); + if (explicit_receiver) PUSH_INSN(ret, location, dup); PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_METHOD), rb_id2sym(method_id), PUSH_VAL(DEFINED_METHOD)); } else { + ID method_id = pm_constant_id_lookup(scope_node, cast->name); + PUSH_INSN(ret, location, putself); if (explicit_receiver) PUSH_INSN(ret, location, dup); + PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_FUNC), rb_id2sym(method_id), PUSH_VAL(DEFINED_METHOD)); } return; - } - case PM_YIELD_NODE: - PUSH_INSN(ret, location, putnil); - PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_YIELD), 0, PUSH_VAL(DEFINED_YIELD)); - return; - case PM_SUPER_NODE: - case PM_FORWARDING_SUPER_NODE: - PUSH_INSN(ret, location, putnil); - PUSH_INSN3(ret, location, defined, INT2FIX(DEFINED_ZSUPER), 0, PUSH_VAL(DEFINED_ZSUPER)); - return; - case PM_CALL_AND_WRITE_NODE: - case PM_CALL_OPERATOR_WRITE_NODE: - case PM_CALL_OR_WRITE_NODE: - - case PM_CONSTANT_WRITE_NODE: - case PM_CONSTANT_OPERATOR_WRITE_NODE: - case PM_CONSTANT_AND_WRITE_NODE: - case PM_CONSTANT_OR_WRITE_NODE: - case PM_CONSTANT_PATH_AND_WRITE_NODE: - case PM_CONSTANT_PATH_OPERATOR_WRITE_NODE: - case PM_CONSTANT_PATH_OR_WRITE_NODE: - case PM_CONSTANT_PATH_WRITE_NODE: +#undef BLOCK_P + } + case PM_ARGUMENTS_NODE: { + // defined?(a(1, 2, 3)) + // ^^^^^^^ + const pm_arguments_node_t *cast = (const pm_arguments_node_t *) node; - case PM_GLOBAL_VARIABLE_WRITE_NODE: - case PM_GLOBAL_VARIABLE_OPERATOR_WRITE_NODE: - case PM_GLOBAL_VARIABLE_AND_WRITE_NODE: - case PM_GLOBAL_VARIABLE_OR_WRITE_NODE: + for (size_t index = 0; index < cast->arguments.size; index++) { + pm_compile_defined_expr0(iseq, cast->arguments.nodes[index], node_location, ret, popped, scope_node, in_condition, lfinish, false); + PUSH_INSNL(ret, location, branchunless, lfinish[1]); + } - case PM_CLASS_VARIABLE_WRITE_NODE: - case PM_CLASS_VARIABLE_OPERATOR_WRITE_NODE: - case PM_CLASS_VARIABLE_AND_WRITE_NODE: - case PM_CLASS_VARIABLE_OR_WRITE_NODE: + dtype = DEFINED_EXPR; + break; + } + case PM_BLOCK_ARGUMENT_NODE: + // defined?(a(&b)) + // ^^ + dtype = DEFINED_EXPR; + break; + case PM_FORWARDING_ARGUMENTS_NODE: + // def a(...) = defined?(a(...)) + // ^^^ + dtype = DEFINED_EXPR; + break; + case PM_SPLAT_NODE: { + // def a(*) = defined?(a(*)) + // ^ + const pm_splat_node_t *cast = (const pm_splat_node_t *) node; - case PM_INDEX_AND_WRITE_NODE: - case PM_INDEX_OPERATOR_WRITE_NODE: - case PM_INDEX_OR_WRITE_NODE: + if (cast->expression == NULL) { + dtype = DEFINED_EXPR; + break; + } - case PM_INSTANCE_VARIABLE_WRITE_NODE: - case PM_INSTANCE_VARIABLE_OPERATOR_WRITE_NODE: - case PM_INSTANCE_VARIABLE_AND_WRITE_NODE: - case PM_INSTANCE_VARIABLE_OR_WRITE_NODE: + pm_compile_defined_expr0(iseq, cast->expression, node_location, ret, popped, scope_node, in_condition, lfinish, false); - case PM_LOCAL_VARIABLE_WRITE_NODE: - case PM_LOCAL_VARIABLE_OPERATOR_WRITE_NODE: - case PM_LOCAL_VARIABLE_AND_WRITE_NODE: - case PM_LOCAL_VARIABLE_OR_WRITE_NODE: + if (!lfinish[1]) lfinish[1] = NEW_LABEL(location.line); + PUSH_INSNL(ret, location, branchunless, lfinish[1]); - case PM_MULTI_WRITE_NODE: - dtype = DEFINED_ASGN; + dtype = DEFINED_EXPR; break; - default: - rb_bug("Unsupported node %s", pm_node_type_to_str(PM_NODE_TYPE(node))); + } + case PM_SHAREABLE_CONSTANT_NODE: + // # shareable_constant_value: literal + // defined?(A = 1) + // ^^^^^ + pm_compile_defined_expr0(iseq, ((const pm_shareable_constant_node_t *) node)->write, node_location, ret, popped, scope_node, in_condition, lfinish, explicit_receiver); + return; +/* Unreachable (parameters) ***************************************************/ + case PM_BLOCK_LOCAL_VARIABLE_NODE: + case PM_BLOCK_PARAMETER_NODE: + case PM_BLOCK_PARAMETERS_NODE: + case PM_FORWARDING_PARAMETER_NODE: + case PM_IMPLICIT_REST_NODE: + case PM_IT_PARAMETERS_NODE: + case PM_PARAMETERS_NODE: + case PM_KEYWORD_REST_PARAMETER_NODE: + case PM_NO_KEYWORDS_PARAMETER_NODE: + case PM_NO_BLOCK_PARAMETER_NODE: + case PM_NUMBERED_PARAMETERS_NODE: + case PM_OPTIONAL_KEYWORD_PARAMETER_NODE: + case PM_OPTIONAL_PARAMETER_NODE: + case PM_REQUIRED_KEYWORD_PARAMETER_NODE: + case PM_REQUIRED_PARAMETER_NODE: + case PM_REST_PARAMETER_NODE: +/* Unreachable (pattern matching) *********************************************/ + case PM_ALTERNATION_PATTERN_NODE: + case PM_ARRAY_PATTERN_NODE: + case PM_CAPTURE_PATTERN_NODE: + case PM_FIND_PATTERN_NODE: + case PM_HASH_PATTERN_NODE: + case PM_PINNED_EXPRESSION_NODE: + case PM_PINNED_VARIABLE_NODE: +/* Unreachable (indirect writes) **********************************************/ + case PM_CALL_TARGET_NODE: + case PM_CLASS_VARIABLE_TARGET_NODE: + case PM_CONSTANT_PATH_TARGET_NODE: + case PM_CONSTANT_TARGET_NODE: + case PM_GLOBAL_VARIABLE_TARGET_NODE: + case PM_INDEX_TARGET_NODE: + case PM_INSTANCE_VARIABLE_TARGET_NODE: + case PM_LOCAL_VARIABLE_TARGET_NODE: + case PM_MULTI_TARGET_NODE: +/* Unreachable (clauses) ******************************************************/ + case PM_ELSE_NODE: + case PM_ENSURE_NODE: + case PM_IN_NODE: + case PM_RESCUE_NODE: + case PM_WHEN_NODE: +/* Unreachable (miscellaneous) ************************************************/ + case PM_BLOCK_NODE: + case PM_EMBEDDED_STATEMENTS_NODE: + case PM_EMBEDDED_VARIABLE_NODE: + case PM_ERROR_RECOVERY_NODE: + case PM_PRE_EXECUTION_NODE: + case PM_PROGRAM_NODE: + case PM_SCOPE_NODE: + case PM_STATEMENTS_NODE: + rb_bug("Unreachable node in defined?: %s", pm_node_type(PM_NODE_TYPE(node))); } RUBY_ASSERT(dtype != DEFINED_NOT_DEFINED); PUSH_INSN1(ret, location, putobject, PUSH_VAL(dtype)); + #undef PUSH_VAL } static void -pm_defined_expr(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_location_t *node_location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node, bool in_condition, LABEL **lfinish, bool explicit_receiver) +pm_defined_expr(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_location_t *node_location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node, bool in_condition, LABEL **lfinish) { LINK_ELEMENT *lcur = ret->last; pm_compile_defined_expr0(iseq, node, node_location, ret, popped, scope_node, in_condition, lfinish, false); @@ -3923,7 +4711,7 @@ pm_compile_defined_expr(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_lo lfinish[2] = 0; if (!popped) { - pm_defined_expr(iseq, node, node_location, ret, popped, scope_node, in_condition, lfinish, false); + pm_defined_expr(iseq, node, node_location, ret, popped, scope_node, in_condition, lfinish); } if (lfinish[1]) { @@ -3951,13 +4739,11 @@ pm_add_ensure_iseq(LINK_ANCHOR *const ret, rb_iseq_t *iseq, int is_return, pm_sc struct iseq_compile_data_ensure_node_stack *prev_enlp = enlp; DECL_ANCHOR(ensure); - INIT_ANCHOR(ensure); while (enlp) { if (enlp->erange != NULL) { DECL_ANCHOR(ensure_part); LABEL *lstart = NEW_LABEL(0); LABEL *lend = NEW_LABEL(0); - INIT_ANCHOR(ensure_part); add_ensure_range(iseq, enlp->erange, lstart, lend); @@ -3979,33 +4765,7 @@ pm_add_ensure_iseq(LINK_ANCHOR *const ret, rb_iseq_t *iseq, int is_return, pm_sc PUSH_SEQ(ret, ensure); } -struct pm_local_table_insert_ctx { - pm_scope_node_t *scope_node; - rb_ast_id_table_t *local_table_for_iseq; - int local_index; -}; - -static int -pm_local_table_insert_func(st_data_t *key, st_data_t *value, st_data_t arg, int existing) -{ - if (!existing) { - pm_constant_id_t constant_id = (pm_constant_id_t) *key; - struct pm_local_table_insert_ctx * ctx = (struct pm_local_table_insert_ctx *) arg; - - pm_scope_node_t *scope_node = ctx->scope_node; - rb_ast_id_table_t *local_table_for_iseq = ctx->local_table_for_iseq; - int local_index = ctx->local_index; - - ID local = pm_constant_id_lookup(scope_node, constant_id); - local_table_for_iseq->ids[local_index] = local; - - *value = (st_data_t)local_index; - - ctx->local_index++; - } - return ST_CONTINUE; -} /** * Insert a local into the local table for the iseq. This is used to create the @@ -4013,24 +4773,23 @@ pm_local_table_insert_func(st_data_t *key, st_data_t *value, st_data_t arg, int * inserted are regular named locals, as opposed to special forwarding locals. */ static void -pm_insert_local_index(pm_constant_id_t constant_id, int local_index, st_table *index_lookup_table, rb_ast_id_table_t *local_table_for_iseq, pm_scope_node_t *scope_node) +pm_insert_local_index(pm_constant_id_t constant_id, int local_index, pm_index_lookup_table_t *index_lookup_table, rb_ast_id_table_t *local_table_for_iseq, pm_scope_node_t *scope_node) { RUBY_ASSERT((constant_id & PM_SPECIAL_CONSTANT_FLAG) == 0); ID local = pm_constant_id_lookup(scope_node, constant_id); local_table_for_iseq->ids[local_index] = local; - st_insert(index_lookup_table, (st_data_t) constant_id, (st_data_t) local_index); + pm_index_lookup_table_insert(index_lookup_table, constant_id, local_index); } /** - * Insert a local into the local table for the iseq that is a special forwarding - * local variable. + * Insert a special forwarding local (*, **, &, ...) into the local table. */ static void -pm_insert_local_special(ID local_name, int local_index, st_table *index_lookup_table, rb_ast_id_table_t *local_table_for_iseq) +pm_insert_local_special(pm_constant_id_t special_id, ID local_name, int local_index, pm_index_lookup_table_t *index_lookup_table, rb_ast_id_table_t *local_table_for_iseq) { local_table_for_iseq->ids[local_index] = local_name; - st_insert(index_lookup_table, (st_data_t) (local_name | PM_SPECIAL_CONSTANT_FLAG), (st_data_t) local_index); + pm_index_lookup_table_insert(index_lookup_table, special_id, local_index); } /** @@ -4040,7 +4799,7 @@ pm_insert_local_special(ID local_name, int local_index, st_table *index_lookup_t * local and index lookup tables and increments the local index as necessary. */ static int -pm_compile_destructured_param_locals(const pm_multi_target_node_t *node, st_table *index_lookup_table, rb_ast_id_table_t *local_table_for_iseq, pm_scope_node_t *scope_node, int local_index) +pm_compile_destructured_param_locals(const pm_multi_target_node_t *node, pm_index_lookup_table_t *index_lookup_table, rb_ast_id_table_t *local_table_for_iseq, pm_scope_node_t *scope_node, int local_index) { for (size_t index = 0; index < node->lefts.size; index++) { const pm_node_t *left = node->lefts.nodes[index]; @@ -4093,9 +4852,9 @@ pm_compile_destructured_param_locals(const pm_multi_target_node_t *node, st_tabl * as a positional parameter in a method, block, or lambda definition. */ static inline void -pm_compile_destructured_param_write(rb_iseq_t *iseq, const pm_required_parameter_node_t *node, LINK_ANCHOR *const ret, const pm_scope_node_t *scope_node) +pm_compile_destructured_param_write(rb_iseq_t *iseq, const pm_required_parameter_node_t *node, LINK_ANCHOR *const ret, pm_scope_node_t *scope_node) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); pm_local_index_t index = pm_lookup_local_index(iseq, scope_node, node->name, 0); PUSH_SETLOCAL(ret, location, index.index, index.level); } @@ -4109,9 +4868,9 @@ pm_compile_destructured_param_write(rb_iseq_t *iseq, const pm_required_parameter * for this simplified case. */ static void -pm_compile_destructured_param_writes(rb_iseq_t *iseq, const pm_multi_target_node_t *node, LINK_ANCHOR *const ret, const pm_scope_node_t *scope_node) +pm_compile_destructured_param_writes(rb_iseq_t *iseq, const pm_multi_target_node_t *node, LINK_ANCHOR *const ret, pm_scope_node_t *scope_node) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); bool has_rest = (node->rest && PM_NODE_TYPE_P(node->rest, PM_SPLAT_NODE) && (((const pm_splat_node_t *) node->rest)->expression) != NULL); bool has_rights = node->rights.size > 0; @@ -4270,7 +5029,7 @@ pm_multi_target_state_update(pm_multi_target_state_t *state) previous = current; current = current->next; - xfree(previous); + SIZED_FREE(previous); } } @@ -4308,7 +5067,7 @@ pm_compile_multi_target_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR static void pm_compile_target_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const parents, LINK_ANCHOR *const writes, LINK_ANCHOR *const cleanup, pm_scope_node_t *scope_node, pm_multi_target_state_t *state) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); switch (PM_NODE_TYPE(node)) { case PM_LOCAL_VARIABLE_TARGET_NODE: { @@ -4332,7 +5091,8 @@ pm_compile_target_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *cons const pm_class_variable_target_node_t *cast = (const pm_class_variable_target_node_t *) node; ID name = pm_constant_id_lookup(scope_node, cast->name); - PUSH_INSN2(writes, location, setclassvariable, ID2SYM(name), get_cvar_ic_value(iseq, name)); + VALUE operand = ID2SYM(name); + PUSH_INSN2(writes, location, setclassvariable, operand, get_cvar_ic_value(iseq, name)); break; } case PM_CONSTANT_TARGET_NODE: { @@ -4344,8 +5104,9 @@ pm_compile_target_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *cons const pm_constant_target_node_t *cast = (const pm_constant_target_node_t *) node; ID name = pm_constant_id_lookup(scope_node, cast->name); + VALUE operand = ID2SYM(name); PUSH_INSN1(writes, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_CONST_BASE)); - PUSH_INSN1(writes, location, setconstant, ID2SYM(name)); + PUSH_INSN1(writes, location, setconstant, operand); break; } case PM_GLOBAL_VARIABLE_TARGET_NODE: { @@ -4357,7 +5118,8 @@ pm_compile_target_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *cons const pm_global_variable_target_node_t *cast = (const pm_global_variable_target_node_t *) node; ID name = pm_constant_id_lookup(scope_node, cast->name); - PUSH_INSN1(writes, location, setglobal, ID2SYM(name)); + VALUE operand = ID2SYM(name); + PUSH_INSN1(writes, location, setglobal, operand); break; } case PM_INSTANCE_VARIABLE_TARGET_NODE: { @@ -4369,7 +5131,8 @@ pm_compile_target_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *cons const pm_instance_variable_target_node_t *cast = (const pm_instance_variable_target_node_t *) node; ID name = pm_constant_id_lookup(scope_node, cast->name); - PUSH_INSN2(writes, location, setinstancevariable, ID2SYM(name), get_ivar_ic_value(iseq, name)); + VALUE operand = ID2SYM(name); + PUSH_INSN2(writes, location, setinstancevariable, operand, get_ivar_ic_value(iseq, name)); break; } case PM_CONSTANT_PATH_TARGET_NODE: { @@ -4399,7 +5162,8 @@ pm_compile_target_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *cons pm_multi_target_state_push(state, (INSN *) LAST_ELEMENT(writes), 1); } - PUSH_INSN1(writes, location, setconstant, ID2SYM(name)); + VALUE operand = ID2SYM(name); + PUSH_INSN1(writes, location, setconstant, operand); if (state != NULL) { PUSH_INSN(cleanup, location, pop); @@ -4464,7 +5228,7 @@ pm_compile_target_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *cons int flags = 0; struct rb_callinfo_kwarg *kwargs = NULL; - int argc = pm_setup_args(cast->arguments, cast->block, &flags, &kwargs, iseq, parents, scope_node, &location); + int argc = pm_setup_args(cast->arguments, (const pm_node_t *) cast->block, &flags, &kwargs, iseq, parents, scope_node, &location); if (state != NULL) { PUSH_INSN1(writes, location, topn, INT2FIX(argc + 1)); @@ -4525,8 +5289,22 @@ pm_compile_target_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *cons break; } + case PM_SPLAT_NODE: { + // Splat nodes capture all values into an array. They can be used + // as targets in assignments or for loops. + // + // for *x in []; end + // + const pm_splat_node_t *cast = (const pm_splat_node_t *) node; + + if (cast->expression != NULL) { + pm_compile_target_node(iseq, cast->expression, parents, writes, cleanup, scope_node, state); + } + + break; + } default: - rb_bug("Unexpected node type: %s", pm_node_type_to_str(PM_NODE_TYPE(node))); + rb_bug("Unexpected node type: %s", pm_node_type(PM_NODE_TYPE(node))); break; } } @@ -4539,7 +5317,7 @@ pm_compile_target_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *cons static void pm_compile_multi_target_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const parents, LINK_ANCHOR *const writes, LINK_ANCHOR *const cleanup, pm_scope_node_t *scope_node, pm_multi_target_state_t *state) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); const pm_node_list_t *lefts; const pm_node_t *rest; const pm_node_list_t *rights; @@ -4560,7 +5338,7 @@ pm_compile_multi_target_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR break; } default: - rb_bug("Unsupported node %s", pm_node_type_to_str(PM_NODE_TYPE(node))); + rb_bug("Unsupported node %s", pm_node_type(PM_NODE_TYPE(node))); break; } @@ -4622,7 +5400,7 @@ pm_compile_multi_target_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR static void pm_compile_for_node_index(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, pm_scope_node_t *scope_node) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); switch (PM_NODE_TYPE(node)) { case PM_LOCAL_VARIABLE_TARGET_NODE: { @@ -4643,10 +5421,7 @@ pm_compile_for_node_index(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *c // owning expression of this target, then retrieve the value, expand it, // and then compile the necessary writes. DECL_ANCHOR(writes); - INIT_ANCHOR(writes); - DECL_ANCHOR(cleanup); - INIT_ANCHOR(cleanup); pm_multi_target_state_t state = { 0 }; state.position = 1; @@ -4661,12 +5436,10 @@ pm_compile_for_node_index(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *c pm_multi_target_state_update(&state); break; } + case PM_SPLAT_NODE: case PM_MULTI_TARGET_NODE: { DECL_ANCHOR(writes); - INIT_ANCHOR(writes); - DECL_ANCHOR(cleanup); - INIT_ANCHOR(cleanup); pm_compile_target_node(iseq, node, ret, writes, cleanup, scope_node, NULL); @@ -4699,12 +5472,18 @@ pm_compile_for_node_index(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *c PUSH_INSN(ret, location, pop); PUSH_LABEL(ret, not_single); + + if (PM_NODE_TYPE_P(node, PM_SPLAT_NODE)) { + const pm_splat_node_t *cast = (const pm_splat_node_t *) node; + PUSH_INSN2(ret, location, expandarray, INT2FIX(0), INT2FIX(cast->expression == NULL ? 0 : 1)); + } + PUSH_SEQ(ret, writes); PUSH_SEQ(ret, cleanup); break; } default: - rb_bug("Unexpected node type for index in for node: %s", pm_node_type_to_str(PM_NODE_TYPE(node))); + rb_bug("Unexpected node type for index in for node: %s", pm_node_type(PM_NODE_TYPE(node))); break; } } @@ -4712,8 +5491,6 @@ pm_compile_for_node_index(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *c static void pm_compile_rescue(rb_iseq_t *iseq, const pm_begin_node_t *cast, const pm_node_location_t *node_location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) { - const pm_parser_t *parser = scope_node->parser; - LABEL *lstart = NEW_LABEL(node_location->line); LABEL *lend = NEW_LABEL(node_location->line); LABEL *lcont = NEW_LABEL(node_location->line); @@ -4725,7 +5502,7 @@ pm_compile_rescue(rb_iseq_t *iseq, const pm_begin_node_t *cast, const pm_node_lo &rescue_scope_node, rb_str_concat(rb_str_new2("rescue in "), ISEQ_BODY(iseq)->location.label), ISEQ_TYPE_RESCUE, - pm_node_line_number(parser, (const pm_node_t *) cast->rescue_clause) + pm_node_line_number_cached((const pm_node_t *) cast->rescue_clause, scope_node) ); pm_scope_node_destroy(&rescue_scope_node); @@ -4741,7 +5518,7 @@ pm_compile_rescue(rb_iseq_t *iseq, const pm_begin_node_t *cast, const pm_node_lo PM_COMPILE_NOT_POPPED((const pm_node_t *) cast->statements); } else { - const pm_node_location_t location = PM_NODE_START_LOCATION(parser, cast->rescue_clause); + const pm_node_location_t location = PM_NODE_START_LOCATION(cast->rescue_clause); PUSH_INSN(ret, location, putnil); } @@ -4764,12 +5541,11 @@ pm_compile_rescue(rb_iseq_t *iseq, const pm_begin_node_t *cast, const pm_node_lo static void pm_compile_ensure(rb_iseq_t *iseq, const pm_begin_node_t *cast, const pm_node_location_t *node_location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) { - const pm_parser_t *parser = scope_node->parser; const pm_statements_node_t *statements = cast->ensure_clause->statements; pm_node_location_t location; if (statements != NULL) { - location = PM_NODE_START_LOCATION(parser, statements); + location = PM_NODE_START_LOCATION(statements); } else { location = *node_location; @@ -4784,7 +5560,6 @@ pm_compile_ensure(rb_iseq_t *iseq, const pm_begin_node_t *cast, const pm_node_lo struct ensure_range *erange; DECL_ANCHOR(ensr); - INIT_ANCHOR(ensr); if (statements != NULL) { pm_compile_node(iseq, (const pm_node_t *) statements, ensr, true, scope_node); } @@ -4854,55 +5629,18 @@ pm_opt_str_freeze_p(const rb_iseq_t *iseq, const pm_call_node_t *node) } /** - * Returns true if the given call node can use the opt_aref_with optimization - * with the current iseq options. - */ -static inline bool -pm_opt_aref_with_p(const rb_iseq_t *iseq, const pm_call_node_t *node) -{ - return ( - !PM_NODE_FLAG_P(node, PM_CALL_NODE_FLAGS_SAFE_NAVIGATION) && - node->arguments != NULL && - PM_NODE_TYPE_P((const pm_node_t *) node->arguments, PM_ARGUMENTS_NODE) && - ((const pm_arguments_node_t *) node->arguments)->arguments.size == 1 && - PM_NODE_TYPE_P(((const pm_arguments_node_t *) node->arguments)->arguments.nodes[0], PM_STRING_NODE) && - node->block == NULL && - !PM_NODE_FLAG_P(((const pm_arguments_node_t *) node->arguments)->arguments.nodes[0], PM_STRING_FLAGS_FROZEN) && - ISEQ_COMPILE_DATA(iseq)->option->specialized_instruction - ); -} - -/** - * Returns true if the given call node can use the opt_aset_with optimization - * with the current iseq options. - */ -static inline bool -pm_opt_aset_with_p(const rb_iseq_t *iseq, const pm_call_node_t *node) -{ - return ( - !PM_NODE_FLAG_P(node, PM_CALL_NODE_FLAGS_SAFE_NAVIGATION) && - node->arguments != NULL && - PM_NODE_TYPE_P((const pm_node_t *) node->arguments, PM_ARGUMENTS_NODE) && - ((const pm_arguments_node_t *) node->arguments)->arguments.size == 2 && - PM_NODE_TYPE_P(((const pm_arguments_node_t *) node->arguments)->arguments.nodes[0], PM_STRING_NODE) && - node->block == NULL && - !PM_NODE_FLAG_P(((const pm_arguments_node_t *) node->arguments)->arguments.nodes[0], PM_STRING_FLAGS_FROZEN) && - ISEQ_COMPILE_DATA(iseq)->option->specialized_instruction - ); -} - -/** * Compile the instructions necessary to read a constant, based on the options * of the current iseq. */ static void -pm_compile_constant_read(rb_iseq_t *iseq, VALUE name, const pm_location_t *name_loc, uint32_t node_id, LINK_ANCHOR *const ret, const pm_scope_node_t *scope_node) +pm_compile_constant_read(rb_iseq_t *iseq, VALUE name, const pm_location_t *name_loc, uint32_t node_id, LINK_ANCHOR *const ret, pm_scope_node_t *scope_node) { - const pm_node_location_t location = PM_LOCATION_START_LOCATION(scope_node->parser, name_loc, node_id); + const pm_node_location_t location = PM_LOCATION_START_LOCATION(name_loc, node_id); if (ISEQ_COMPILE_DATA(iseq)->option->inline_const_cache) { ISEQ_BODY(iseq)->ic_size++; VALUE segments = rb_ary_new_from_args(1, name); + RB_OBJ_SET_SHAREABLE(segments); PUSH_INSN1(ret, location, opt_getconstant_path, segments); } else { @@ -4957,7 +5695,7 @@ pm_constant_path_parts(const pm_node_t *node, const pm_scope_node_t *scope_node) static void pm_compile_constant_path(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const prefix, LINK_ANCHOR *const body, bool popped, pm_scope_node_t *scope_node) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); switch (PM_NODE_TYPE(node)) { case PM_CONSTANT_READ_NODE: { @@ -4992,67 +5730,10 @@ pm_compile_constant_path(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *co } /** - * When we're compiling a case node, it's possible that we can speed it up using - * a dispatch hash, which will allow us to jump directly to the correct when - * clause body based on a hash lookup of the value. This can only happen when - * the conditions are literals that can be compiled into a hash key. - * - * This function accepts a dispatch hash and the condition of a when clause. It - * is responsible for compiling the condition into a hash key and then adding it - * to the dispatch hash. - * - * If the value can be successfully compiled into the hash, then this function - * returns the dispatch hash with the new key added. If the value cannot be - * compiled into the hash, then this function returns Qundef. In the case of - * Qundef, this function is signaling that the caller should abandon the - * optimization entirely. - */ -static VALUE -pm_compile_case_node_dispatch(rb_iseq_t *iseq, VALUE dispatch, const pm_node_t *node, LABEL *label, const pm_scope_node_t *scope_node) -{ - VALUE key = Qundef; - - switch (PM_NODE_TYPE(node)) { - case PM_FLOAT_NODE: { - key = pm_static_literal_value(iseq, node, scope_node); - double intptr; - - if (modf(RFLOAT_VALUE(key), &intptr) == 0.0) { - key = (FIXABLE(intptr) ? LONG2FIX((long) intptr) : rb_dbl2big(intptr)); - } - - break; - } - case PM_FALSE_NODE: - case PM_INTEGER_NODE: - case PM_NIL_NODE: - case PM_SOURCE_FILE_NODE: - case PM_SOURCE_LINE_NODE: - case PM_SYMBOL_NODE: - case PM_TRUE_NODE: - key = pm_static_literal_value(iseq, node, scope_node); - break; - case PM_STRING_NODE: { - const pm_string_node_t *cast = (const pm_string_node_t *) node; - key = parse_static_literal_string(iseq, scope_node, node, &cast->unescaped); - break; - } - default: - return Qundef; - } - - if (NIL_P(rb_hash_lookup(dispatch, key))) { - rb_hash_aset(dispatch, key, ((VALUE) label) | 1); - } - - return dispatch; -} - -/** * Return the object that will be pushed onto the stack for the given node. */ static VALUE -pm_compile_shareable_constant_literal(rb_iseq_t *iseq, const pm_node_t *node, const pm_scope_node_t *scope_node) +pm_compile_shareable_constant_literal(rb_iseq_t *iseq, const pm_node_t *node, pm_scope_node_t *scope_node) { switch (PM_NODE_TYPE(node)) { case PM_TRUE_NODE: @@ -5086,7 +5767,7 @@ pm_compile_shareable_constant_literal(rb_iseq_t *iseq, const pm_node_t *node, co } case PM_HASH_NODE: { const pm_hash_node_t *cast = (const pm_hash_node_t *) node; - VALUE result = rb_hash_new_capa(cast->elements.size); + VALUE result = rb_hash_alloc_fixed_size(rb_cHash, cast->elements.size); for (size_t index = 0; index < cast->elements.size; index++) { const pm_node_t *element = cast->elements.nodes[index]; @@ -5119,12 +5800,12 @@ pm_compile_shareable_constant_value(rb_iseq_t *iseq, const pm_node_t *node, cons { VALUE literal = pm_compile_shareable_constant_literal(iseq, node, scope_node); if (literal != Qundef) { - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); PUSH_INSN1(ret, location, putobject, literal); return; } - const pm_node_location_t location = PM_NODE_START_LOCATION(scope_node->parser, node); + const pm_node_location_t location = PM_NODE_START_LOCATION(node); switch (PM_NODE_TYPE(node)) { case PM_ARRAY_NODE: { const pm_array_node_t *cast = (const pm_array_node_t *) node; @@ -5153,19 +5834,7 @@ pm_compile_shareable_constant_value(rb_iseq_t *iseq, const pm_node_t *node, cons PUSH_INSN1(ret, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); } - for (size_t index = 0; index < cast->elements.size; index++) { - const pm_node_t *element = cast->elements.nodes[index]; - - if (!PM_NODE_TYPE_P(element, PM_ASSOC_NODE)) { - COMPILE_ERROR(iseq, location.line, "Ractor constant writes do not support **"); - } - - const pm_assoc_node_t *assoc = (const pm_assoc_node_t *) element; - pm_compile_shareable_constant_value(iseq, assoc->key, shareability, path, ret, scope_node, false); - pm_compile_shareable_constant_value(iseq, assoc->value, shareability, path, ret, scope_node, false); - } - - PUSH_INSN1(ret, location, newhash, INT2FIX(cast->elements.size * 2)); + pm_compile_hash_elements(iseq, (const pm_node_t *) cast, &cast->elements, shareability, path, false, ret, scope_node); if (top) { ID method_id = (shareability & PM_SHAREABLE_CONSTANT_NODE_FLAGS_EXPERIMENTAL_COPY) ? rb_intern("make_shareable_copy") : rb_intern("make_shareable"); @@ -5176,7 +5845,6 @@ pm_compile_shareable_constant_value(rb_iseq_t *iseq, const pm_node_t *node, cons } default: { DECL_ANCHOR(value_seq); - INIT_ANCHOR(value_seq); pm_compile_node(iseq, node, value_seq, false, scope_node); if (PM_NODE_TYPE_P(node, PM_INTERPOLATED_STRING_NODE)) { @@ -5186,6 +5854,9 @@ pm_compile_shareable_constant_value(rb_iseq_t *iseq, const pm_node_t *node, cons if (shareability & PM_SHAREABLE_CONSTANT_NODE_FLAGS_LITERAL) { PUSH_INSN1(ret, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); PUSH_SEQ(ret, value_seq); + if (!RB_OBJ_SHAREABLE_P(path)) { + RB_OBJ_SET_SHAREABLE(path); + } PUSH_INSN1(ret, location, putobject, path); PUSH_SEND_WITH_FLAG(ret, location, rb_intern("ensure_shareable"), INT2FIX(2), INT2FIX(VM_CALL_ARGS_SIMPLE)); } @@ -5224,7 +5895,9 @@ pm_compile_constant_write_node(rb_iseq_t *iseq, const pm_constant_write_node_t * if (!popped) PUSH_INSN(ret, location, dup); PUSH_INSN1(ret, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_CONST_BASE)); - PUSH_INSN1(ret, location, setconstant, ID2SYM(name_id)); + + VALUE operand = ID2SYM(name_id); + PUSH_INSN1(ret, location, setconstant, operand); } /** @@ -5538,1021 +6211,2718 @@ pm_compile_constant_path_operator_write_node(rb_iseq_t *iseq, const pm_constant_ } /** - * Compiles a prism node into instruction sequences. - * - * iseq - The current instruction sequence object (used for locals) - * node - The prism node to compile - * ret - The linked list of instructions to append instructions onto - * popped - True if compiling something with no side effects, so instructions don't - * need to be added - * scope_node - Stores parser and local information + * Many nodes in Prism can be marked as a static literal, which means slightly + * different things depending on which node it is. Occasionally we need to omit + * container nodes from static literal checks, which is where this macro comes + * in. */ -static void -pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +#define PM_CONTAINER_P(node) (PM_NODE_TYPE_P(node, PM_ARRAY_NODE) || PM_NODE_TYPE_P(node, PM_HASH_NODE) || PM_NODE_TYPE_P(node, PM_RANGE_NODE)) + +/** + * Compile a scope node, which is a special kind of node that represents a new + * lexical scope, attached to a node in the AST. + */ +static inline void +pm_compile_scope_node(rb_iseq_t *iseq, pm_scope_node_t *scope_node, const pm_node_location_t *node_location, LINK_ANCHOR *const ret, bool popped) { - const pm_parser_t *parser = scope_node->parser; - const pm_node_location_t location = PM_NODE_START_LOCATION(parser, node); - int lineno = (int) location.line; + const pm_node_location_t location = *node_location; + struct rb_iseq_constant_body *body = ISEQ_BODY(iseq); - if (PM_NODE_TYPE_P(node, PM_BEGIN_NODE) && (((const pm_begin_node_t *) node)->statements == NULL) && (((const pm_begin_node_t *) node)->rescue_clause != NULL)) { - // If this node is a begin node and it has empty statements and also - // has a rescue clause, then the other parser considers it as - // starting on the same line as the rescue, as opposed to the - // location of the begin keyword. We replicate that behavior here. - lineno = (int) PM_NODE_START_LINE_COLUMN(parser, ((const pm_begin_node_t *) node)->rescue_clause).line; + pm_constant_id_list_t *locals = &scope_node->locals; + pm_parameters_node_t *parameters_node = NULL; + pm_node_list_t *keywords_list = NULL; + pm_node_list_t *optionals_list = NULL; + pm_node_list_t *posts_list = NULL; + pm_node_list_t *requireds_list = NULL; + pm_node_list_t *block_locals = NULL; + bool trailing_comma = false; + + if (PM_NODE_TYPE_P(scope_node->ast_node, PM_CLASS_NODE) || PM_NODE_TYPE_P(scope_node->ast_node, PM_MODULE_NODE)) { + PUSH_TRACE(ret, RUBY_EVENT_CLASS); + } + + if (scope_node->parameters != NULL) { + switch (PM_NODE_TYPE(scope_node->parameters)) { + case PM_BLOCK_PARAMETERS_NODE: { + pm_block_parameters_node_t *cast = (pm_block_parameters_node_t *) scope_node->parameters; + parameters_node = cast->parameters; + block_locals = &cast->locals; + + if (parameters_node) { + if (parameters_node->rest && PM_NODE_TYPE_P(parameters_node->rest, PM_IMPLICIT_REST_NODE)) { + trailing_comma = true; + } + } + break; + } + case PM_PARAMETERS_NODE: { + parameters_node = (pm_parameters_node_t *) scope_node->parameters; + break; + } + case PM_NUMBERED_PARAMETERS_NODE: { + uint32_t maximum = ((const pm_numbered_parameters_node_t *) scope_node->parameters)->maximum; + body->param.lead_num = maximum; + body->param.flags.ambiguous_param0 = maximum == 1; + break; + } + case PM_IT_PARAMETERS_NODE: + body->param.lead_num = 1; + body->param.flags.ambiguous_param0 = true; + break; + default: + rb_bug("Unexpected node type for parameters: %s", pm_node_type(PM_NODE_TYPE(scope_node->parameters))); + } } - if (PM_NODE_FLAG_P(node, PM_NODE_FLAG_NEWLINE) && ISEQ_COMPILE_DATA(iseq)->last_line != lineno) { - // If this node has the newline flag set and it is on a new line - // from the previous nodes that have been compiled for this ISEQ, - // then we need to emit a newline event. - int event = RUBY_EVENT_LINE; + struct rb_iseq_param_keyword *keyword = NULL; - ISEQ_COMPILE_DATA(iseq)->last_line = lineno; - if (ISEQ_COVERAGE(iseq) && ISEQ_LINE_COVERAGE(iseq)) { - event |= RUBY_EVENT_COVERAGE_LINE; + if (parameters_node) { + optionals_list = ¶meters_node->optionals; + requireds_list = ¶meters_node->requireds; + keywords_list = ¶meters_node->keywords; + posts_list = ¶meters_node->posts; + } + else if (scope_node->parameters && (PM_NODE_TYPE_P(scope_node->parameters, PM_NUMBERED_PARAMETERS_NODE) || PM_NODE_TYPE_P(scope_node->parameters, PM_IT_PARAMETERS_NODE))) { + body->param.opt_num = 0; + } + else { + body->param.lead_num = 0; + body->param.opt_num = 0; + } + + //********STEP 1********** + // Goal: calculate the table size for the locals, accounting for + // hidden variables and multi target nodes + size_t locals_size = locals->size; + + // Index lookup table buffer size is only the number of the locals. + // We'll initialize it after computing table_size below. + pm_index_lookup_table_t index_lookup_table = PM_INDEX_LOOKUP_TABLE_INIT; + + int table_size = (int) locals_size; + + // For nodes have a hidden iteration variable. We add that to the local + // table size here. + if (PM_NODE_TYPE_P(scope_node->ast_node, PM_FOR_NODE)) table_size++; + + if (keywords_list && keywords_list->size) { + table_size++; + } + + if (requireds_list) { + for (size_t i = 0; i < requireds_list->size; i++) { + // For each MultiTargetNode, we're going to have one + // additional anonymous local not represented in the locals table + // We want to account for this in our table size + pm_node_t *required = requireds_list->nodes[i]; + if (PM_NODE_TYPE_P(required, PM_MULTI_TARGET_NODE)) { + table_size++; + } + else if (PM_NODE_TYPE_P(required, PM_REQUIRED_PARAMETER_NODE)) { + if (PM_NODE_FLAG_P(required, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { + table_size++; + } + } } - PUSH_TRACE(ret, event); } - switch (PM_NODE_TYPE(node)) { - case PM_ALIAS_GLOBAL_VARIABLE_NODE: { - // alias $foo $bar - // ^^^^^^^^^^^^^^^ - const pm_alias_global_variable_node_t *cast = (const pm_alias_global_variable_node_t *) node; - PUSH_INSN1(ret, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); + // If we have the `it` implicit local variable, we need to account for + // it in the local table size. + if (scope_node->parameters != NULL && PM_NODE_TYPE_P(scope_node->parameters, PM_IT_PARAMETERS_NODE)) { + table_size++; + } - const pm_location_t *new_name_loc = &cast->new_name->location; - PUSH_INSN1(ret, location, putobject, ID2SYM(rb_intern3((const char *) new_name_loc->start, new_name_loc->end - new_name_loc->start, scope_node->encoding))); + // Ensure there is enough room in the local table for any + // parameters that have been repeated + // ex: def underscore_parameters(_, _ = 1, _ = 2); _; end + // ^^^^^^^^^^^^ + if (optionals_list && optionals_list->size) { + for (size_t i = 0; i < optionals_list->size; i++) { + pm_node_t * node = optionals_list->nodes[i]; + if (PM_NODE_FLAG_P(node, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { + table_size++; + } + } + } - const pm_location_t *old_name_loc = &cast->old_name->location; - PUSH_INSN1(ret, location, putobject, ID2SYM(rb_intern3((const char *) old_name_loc->start, old_name_loc->end - old_name_loc->start, scope_node->encoding))); + // If we have an anonymous "rest" node, we'll need to increase the local + // table size to take it in to account. + // def m(foo, *, bar) + // ^ + if (parameters_node) { + if (parameters_node->rest) { + if (!(PM_NODE_TYPE_P(parameters_node->rest, PM_IMPLICIT_REST_NODE))) { + if (!((const pm_rest_parameter_node_t *) parameters_node->rest)->name || PM_NODE_FLAG_P(parameters_node->rest, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { + table_size++; + } + } + } - PUSH_SEND(ret, location, id_core_set_variable_alias, INT2FIX(2)); - if (popped) PUSH_INSN(ret, location, pop); + // def foo(_, **_); _; end + // ^^^ + if (parameters_node->keyword_rest) { + // def foo(...); end + // ^^^ + // When we have a `...` as the keyword_rest, it's a forwarding_parameter_node and + // we need to leave space for 4 locals: *, **, &, ... + if (PM_NODE_TYPE_P(parameters_node->keyword_rest, PM_FORWARDING_PARAMETER_NODE)) { + // Only optimize specifically methods like this: `foo(...)` + if (requireds_list->size == 0 && optionals_list->size == 0 && keywords_list->size == 0) { + ISEQ_BODY(iseq)->param.flags.use_block = TRUE; + ISEQ_BODY(iseq)->param.flags.forwardable = TRUE; + table_size += 1; + } + else { + table_size += 4; + } + } + else { + const pm_keyword_rest_parameter_node_t *kw_rest = (const pm_keyword_rest_parameter_node_t *) parameters_node->keyword_rest; - return; - } - case PM_ALIAS_METHOD_NODE: { - // alias foo bar - // ^^^^^^^^^^^^^ - const pm_alias_method_node_t *cast = (const pm_alias_method_node_t *) node; + // If it's anonymous or repeated, then we need to allocate stack space + if (!kw_rest->name || PM_NODE_FLAG_P(kw_rest, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { + table_size++; + } + } + } + } - PUSH_INSN1(ret, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); - PUSH_INSN1(ret, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_CBASE)); - PM_COMPILE_NOT_POPPED(cast->new_name); - PM_COMPILE_NOT_POPPED(cast->old_name); + if (posts_list) { + for (size_t i = 0; i < posts_list->size; i++) { + // For each MultiTargetNode, we're going to have one + // additional anonymous local not represented in the locals table + // We want to account for this in our table size + pm_node_t *required = posts_list->nodes[i]; + if (PM_NODE_TYPE_P(required, PM_MULTI_TARGET_NODE) || PM_NODE_FLAG_P(required, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { + table_size++; + } + } + } - PUSH_SEND(ret, location, id_core_set_method_alias, INT2FIX(3)); - if (popped) PUSH_INSN(ret, location, pop); + if (keywords_list && keywords_list->size) { + for (size_t i = 0; i < keywords_list->size; i++) { + pm_node_t *keyword_parameter_node = keywords_list->nodes[i]; + if (PM_NODE_FLAG_P(keyword_parameter_node, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { + table_size++; + } + } + } - return; - } - case PM_AND_NODE: { - // a and b - // ^^^^^^^ - const pm_and_node_t *cast = (const pm_and_node_t *) node; - LABEL *end_label = NEW_LABEL(lineno); + if (parameters_node && parameters_node->block && PM_NODE_TYPE_P(parameters_node->block, PM_BLOCK_PARAMETER_NODE)) { + const pm_block_parameter_node_t *block_node = (const pm_block_parameter_node_t *) parameters_node->block; - PM_COMPILE_NOT_POPPED(cast->left); - if (!popped) PUSH_INSN(ret, location, dup); - PUSH_INSNL(ret, location, branchunless, end_label); + if (PM_NODE_FLAG_P(block_node, PM_PARAMETER_FLAGS_REPEATED_PARAMETER) || !block_node->name) { + table_size++; + } + } - if (!popped) PUSH_INSN(ret, location, pop); - PM_COMPILE(cast->right); - PUSH_LABEL(ret, end_label); + // We can create local_table_for_iseq with the correct size + VALUE idtmp = 0; + rb_ast_id_table_t *local_table_for_iseq = ALLOCV(idtmp, sizeof(rb_ast_id_table_t) + table_size * sizeof(ID)); + local_table_for_iseq->size = table_size; + + // Init the direct-indexed lookup table. The capacity is based on the + // parser's constant pool size (for regular locals) plus special slots. + pm_index_lookup_table_init(&index_lookup_table, (int) pm_parser_constants_size(scope_node->parser), iseq); + + //********END OF STEP 1********** + + //********STEP 2********** + // Goal: populate iv index table as well as local table, keeping the + // layout of the local table consistent with the layout of the + // stack when calling the method + // + // Do a first pass on all of the parameters, setting their values in + // the local_table_for_iseq, _except_ for Multis who get a hidden + // variable in this step, and will get their names inserted in step 3 + + // local_index is a cursor that keeps track of the current + // index into local_table_for_iseq. The local table is actually a list, + // and the order of that list must match the order of the items pushed + // on the stack. We need to take in to account things pushed on the + // stack that _might not have a name_ (for example array destructuring). + // This index helps us know which item we're dealing with and also give + // those anonymous items temporary names (as below) + int local_index = 0; + + // Here we figure out local table indices and insert them in to the + // index lookup table and local tables. + // + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^^^^^^^^^^^^ + if (requireds_list && requireds_list->size) { + for (size_t i = 0; i < requireds_list->size; i++, local_index++) { + ID local; + + // For each MultiTargetNode, we're going to have one additional + // anonymous local not represented in the locals table. We want + // to account for this in our table size. + pm_node_t *required = requireds_list->nodes[i]; + + switch (PM_NODE_TYPE(required)) { + case PM_MULTI_TARGET_NODE: { + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^^^^^^^^^ + local = rb_make_temporary_id(local_index); + local_table_for_iseq->ids[local_index] = local; + break; + } + case PM_REQUIRED_PARAMETER_NODE: { + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^ + const pm_required_parameter_node_t *param = (const pm_required_parameter_node_t *) required; - return; - } - case PM_ARGUMENTS_NODE: - // These are ArgumentsNodes that are not compiled directly by their - // parent call nodes, used in the cases of NextNodes, ReturnNodes, and - // BreakNodes. They can create an array like ArrayNode. - case PM_ARRAY_NODE: { - const pm_node_list_t *elements; + if (PM_NODE_FLAG_P(required, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { + ID local = pm_constant_id_lookup(scope_node, param->name); + local_table_for_iseq->ids[local_index] = local; + } + else { + pm_insert_local_index(param->name, local_index, &index_lookup_table, local_table_for_iseq, scope_node); + } + + break; + } + default: + rb_bug("Unsupported node in requireds in parameters %s", pm_node_type(PM_NODE_TYPE(required))); + } + } - if (PM_NODE_TYPE(node) == PM_ARGUMENTS_NODE) { - // break foo - // ^^^ - const pm_arguments_node_t *cast = (const pm_arguments_node_t *) node; - elements = &cast->arguments; + body->param.lead_num = (int) requireds_list->size; + body->param.flags.has_lead = true; + } - // If we are only returning a single element through one of the jump - // nodes, then we will only compile that node directly. - if (elements->size == 1) { - PM_COMPILE(elements->nodes[0]); - return; + if (scope_node->parameters != NULL && PM_NODE_TYPE_P(scope_node->parameters, PM_IT_PARAMETERS_NODE)) { + local_table_for_iseq->ids[local_index++] = idItImplicit; + } + + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^^^^ + if (optionals_list && optionals_list->size) { + body->param.opt_num = (int) optionals_list->size; + body->param.flags.has_opt = true; + + for (size_t i = 0; i < optionals_list->size; i++, local_index++) { + pm_node_t * node = optionals_list->nodes[i]; + pm_constant_id_t name = ((const pm_optional_parameter_node_t *) node)->name; + + if (PM_NODE_FLAG_P(node, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { + ID local = pm_constant_id_lookup(scope_node, name); + local_table_for_iseq->ids[local_index] = local; + } + else { + pm_insert_local_index(name, local_index, &index_lookup_table, local_table_for_iseq, scope_node); } } - else { - // [foo, bar, baz] - // ^^^^^^^^^^^^^^^ - const pm_array_node_t *cast = (const pm_array_node_t *) node; - elements = &cast->elements; + } + + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^ + if (parameters_node && parameters_node->rest) { + body->param.rest_start = local_index; + + // If there's a trailing comma, we'll have an implicit rest node, + // and we don't want it to impact the rest variables on param + if (!(PM_NODE_TYPE_P(parameters_node->rest, PM_IMPLICIT_REST_NODE))) { + body->param.flags.has_rest = true; + RUBY_ASSERT(body->param.rest_start != -1); + + pm_constant_id_t name = ((const pm_rest_parameter_node_t *) parameters_node->rest)->name; + + if (name) { + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^ + if (PM_NODE_FLAG_P(parameters_node->rest, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { + ID local = pm_constant_id_lookup(scope_node, name); + local_table_for_iseq->ids[local_index] = local; + } + else { + pm_insert_local_index(name, local_index, &index_lookup_table, local_table_for_iseq, scope_node); + } + } + else { + // def foo(a, (b, *c, d), e = 1, *, g, (h, *i, j), k:, l: 1, **m, &n) + // ^ + body->param.flags.anon_rest = true; + pm_insert_local_special(PM_CONSTANT_MULT, idMULT, local_index, &index_lookup_table, local_table_for_iseq); + } + + local_index++; } + } - // If every node in the array is static, then we can compile the entire - // array now instead of later. - if (PM_NODE_FLAG_P(node, PM_NODE_FLAG_STATIC_LITERAL)) { - // We're only going to compile this node if it's not popped. If it - // is popped, then we know we don't need to do anything since it's - // statically known. - if (!popped) { - if (elements->size) { - VALUE value = pm_static_literal_value(iseq, node, scope_node); - PUSH_INSN1(ret, location, duparray, value); + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^^^^^^^^^^^^ + if (posts_list && posts_list->size) { + body->param.post_num = (int) posts_list->size; + body->param.post_start = local_index; + body->param.flags.has_post = true; + + for (size_t i = 0; i < posts_list->size; i++, local_index++) { + ID local; + + // For each MultiTargetNode, we're going to have one additional + // anonymous local not represented in the locals table. We want + // to account for this in our table size. + const pm_node_t *post_node = posts_list->nodes[i]; + + switch (PM_NODE_TYPE(post_node)) { + case PM_MULTI_TARGET_NODE: { + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^^^^^^^^^ + local = rb_make_temporary_id(local_index); + local_table_for_iseq->ids[local_index] = local; + break; + } + case PM_REQUIRED_PARAMETER_NODE: { + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^ + const pm_required_parameter_node_t *param = (const pm_required_parameter_node_t *) post_node; + + if (PM_NODE_FLAG_P(param, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { + ID local = pm_constant_id_lookup(scope_node, param->name); + local_table_for_iseq->ids[local_index] = local; } else { - PUSH_INSN1(ret, location, newarray, INT2FIX(0)); + pm_insert_local_index(param->name, local_index, &index_lookup_table, local_table_for_iseq, scope_node); } + break; + } + default: + rb_bug("Unsupported node in posts in parameters %s", pm_node_type(PM_NODE_TYPE(post_node))); } } - else { - // Here since we know there are possible side-effects inside the - // array contents, we're going to build it entirely at runtime. - // We'll do this by pushing all of the elements onto the stack and - // then combining them with newarray. - // - // If this array is popped, then this serves only to ensure we enact - // all side-effects (like method calls) that are contained within - // the array contents. - // - // We treat all sequences of non-splat elements as their - // own arrays, followed by a newarray, and then continually - // concat the arrays with the SplatNode nodes. - const int max_new_array_size = 0x100; - const unsigned int min_tmp_array_size = 0x40; - - int new_array_size = 0; - bool first_chunk = true; - - // This is an optimization wherein we keep track of whether or not - // the previous element was a static literal. If it was, then we do - // not attempt to check if we have a subarray that can be optimized. - // If it was not, then we do check. - bool static_literal = false; - - // Either create a new array, or push to the existing array. -#define FLUSH_CHUNK \ - if (new_array_size) { \ - if (first_chunk) PUSH_INSN1(ret, location, newarray, INT2FIX(new_array_size)); \ - else PUSH_INSN1(ret, location, pushtoarray, INT2FIX(new_array_size)); \ - first_chunk = false; \ - new_array_size = 0; \ + } + + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^^^^^^^ + // Keywords create an internal variable on the parse tree + if (keywords_list && keywords_list->size) { + keyword = ZALLOC_N(struct rb_iseq_param_keyword, 1); + keyword->num = (int) keywords_list->size; + + const VALUE default_values = rb_ary_hidden_new(1); + const VALUE complex_mark = rb_str_tmp_new(0); + + for (size_t i = 0; i < keywords_list->size; i++) { + pm_node_t *keyword_parameter_node = keywords_list->nodes[i]; + pm_constant_id_t name; + + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^ + if (PM_NODE_TYPE_P(keyword_parameter_node, PM_REQUIRED_KEYWORD_PARAMETER_NODE)) { + name = ((const pm_required_keyword_parameter_node_t *) keyword_parameter_node)->name; + keyword->required_num++; + ID local = pm_constant_id_lookup(scope_node, name); + + if (PM_NODE_FLAG_P(keyword_parameter_node, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { + local_table_for_iseq->ids[local_index] = local; + } + else { + pm_insert_local_index(name, local_index, &index_lookup_table, local_table_for_iseq, scope_node); + } + local_index++; } + } - for (size_t index = 0; index < elements->size; index++) { - const pm_node_t *element = elements->nodes[index]; + for (size_t i = 0; i < keywords_list->size; i++) { + pm_node_t *keyword_parameter_node = keywords_list->nodes[i]; + pm_constant_id_t name; - if (PM_NODE_TYPE_P(element, PM_SPLAT_NODE)) { - FLUSH_CHUNK; + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^^^ + if (PM_NODE_TYPE_P(keyword_parameter_node, PM_OPTIONAL_KEYWORD_PARAMETER_NODE)) { + const pm_optional_keyword_parameter_node_t *cast = ((const pm_optional_keyword_parameter_node_t *) keyword_parameter_node); - const pm_splat_node_t *splat_element = (const pm_splat_node_t *) element; - if (splat_element->expression) { - PM_COMPILE_NOT_POPPED(splat_element->expression); - } - else { - pm_local_index_t index = pm_lookup_local_index(iseq, scope_node, PM_CONSTANT_MULT, 0); - PUSH_GETLOCAL(ret, location, index.index, index.level); - } + pm_node_t *value = cast->value; + name = cast->name; - if (first_chunk) { - // If this is the first element of the array then we - // need to splatarray the elements into the list. - PUSH_INSN1(ret, location, splatarray, Qtrue); - first_chunk = false; - } - else { - PUSH_INSN(ret, location, concattoarray); - } + if (PM_NODE_FLAG_P(value, PM_NODE_FLAG_STATIC_LITERAL) && !PM_CONTAINER_P(value)) { + rb_ary_push(default_values, pm_static_literal_value(iseq, value, scope_node)); + } + else { + rb_ary_push(default_values, complex_mark); + } - static_literal = false; + ID local = pm_constant_id_lookup(scope_node, name); + if (PM_NODE_FLAG_P(keyword_parameter_node, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { + local_table_for_iseq->ids[local_index] = local; } - else if (PM_NODE_TYPE_P(element, PM_KEYWORD_HASH_NODE)) { - if (new_array_size == 0 && first_chunk) { - PUSH_INSN1(ret, location, newarray, INT2FIX(0)); - first_chunk = false; - } - else { - FLUSH_CHUNK; - } + else { + pm_insert_local_index(name, local_index, &index_lookup_table, local_table_for_iseq, scope_node); + } + local_index++; + } - // If we get here, then this is the last element of the - // array/arguments, because it cannot be followed by - // anything else without a syntax error. This looks like: - // - // [foo, bar, baz: qux] - // ^^^^^^^^ - // - // [foo, bar, **baz] - // ^^^^^ - // - const pm_keyword_hash_node_t *keyword_hash = (const pm_keyword_hash_node_t *) element; - pm_compile_hash_elements(iseq, element, &keyword_hash->elements, false, ret, scope_node); - - // This boolean controls the manner in which we push the - // hash onto the array. If it's all keyword splats, then we - // can use the very specialized pushtoarraykwsplat - // instruction to check if it's empty before we push it. - size_t splats = 0; - while (splats < keyword_hash->elements.size && PM_NODE_TYPE_P(keyword_hash->elements.nodes[splats], PM_ASSOC_SPLAT_NODE)) splats++; - - if (keyword_hash->elements.size == splats) { - PUSH_INSN(ret, location, pushtoarraykwsplat); + } + + if (RARRAY_LEN(default_values)) { + VALUE *dvs = ALLOC_N(VALUE, RARRAY_LEN(default_values)); + + for (int i = 0; i < RARRAY_LEN(default_values); i++) { + VALUE dv = RARRAY_AREF(default_values, i); + if (dv == complex_mark) dv = Qundef; + RB_OBJ_WRITE(iseq, &dvs[i], dv); + } + + keyword->default_values = dvs; + } + + // Hidden local for keyword arguments + keyword->bits_start = local_index; + ID local = rb_make_temporary_id(local_index); + local_table_for_iseq->ids[local_index] = local; + local_index++; + + body->param.keyword = keyword; + body->param.flags.has_kw = true; + } + + if (body->type == ISEQ_TYPE_BLOCK && local_index == 1 && requireds_list && requireds_list->size == 1 && !trailing_comma) { + body->param.flags.ambiguous_param0 = true; + } + + if (parameters_node) { + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^^ + if (parameters_node->keyword_rest) { + switch (PM_NODE_TYPE(parameters_node->keyword_rest)) { + case PM_NO_KEYWORDS_PARAMETER_NODE: { + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **nil, &n) + // ^^^^^ + body->param.flags.accepts_no_kwarg = true; + break; + } + case PM_KEYWORD_REST_PARAMETER_NODE: { + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^^ + const pm_keyword_rest_parameter_node_t *kw_rest_node = (const pm_keyword_rest_parameter_node_t *) parameters_node->keyword_rest; + if (!body->param.flags.has_kw) { + body->param.keyword = keyword = ZALLOC_N(struct rb_iseq_param_keyword, 1); + } + + keyword->rest_start = local_index; + body->param.flags.has_kwrest = true; + + pm_constant_id_t constant_id = kw_rest_node->name; + if (constant_id) { + if (PM_NODE_FLAG_P(kw_rest_node, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { + ID local = pm_constant_id_lookup(scope_node, constant_id); + local_table_for_iseq->ids[local_index] = local; } else { - new_array_size++; + pm_insert_local_index(constant_id, local_index, &index_lookup_table, local_table_for_iseq, scope_node); } } - else if (PM_NODE_FLAG_P(element, PM_NODE_FLAG_STATIC_LITERAL) && !static_literal && ((index + min_tmp_array_size) < elements->size)) { - // If we have a static literal, then there's the potential - // to group a bunch of them together with a literal array - // and then concat them together. - size_t right_index = index + 1; - while (right_index < elements->size && PM_NODE_FLAG_P(elements->nodes[right_index], PM_NODE_FLAG_STATIC_LITERAL)) right_index++; - - size_t tmp_array_size = right_index - index; - if (tmp_array_size >= min_tmp_array_size) { - VALUE tmp_array = rb_ary_hidden_new(tmp_array_size); - - // Create the temporary array. - for (; tmp_array_size; tmp_array_size--) - rb_ary_push(tmp_array, pm_static_literal_value(iseq, elements->nodes[index++], scope_node)); - OBJ_FREEZE(tmp_array); - - // Emit the optimized code. - FLUSH_CHUNK; - if (first_chunk) { - PUSH_INSN1(ret, location, duparray, tmp_array); - first_chunk = false; - } - else { - PUSH_INSN1(ret, location, putobject, tmp_array); - PUSH_INSN(ret, location, concattoarray); - } + else { + body->param.flags.anon_kwrest = true; + pm_insert_local_special(PM_CONSTANT_POW, idPow, local_index, &index_lookup_table, local_table_for_iseq); + } + + local_index++; + break; + } + case PM_FORWARDING_PARAMETER_NODE: { + // def foo(...) + // ^^^ + if (!ISEQ_BODY(iseq)->param.flags.forwardable) { + // Add the anonymous * + body->param.rest_start = local_index; + body->param.flags.has_rest = true; + body->param.flags.anon_rest = true; + pm_insert_local_special(PM_CONSTANT_MULT, idMULT, local_index++, &index_lookup_table, local_table_for_iseq); + + // Add the anonymous ** + RUBY_ASSERT(!body->param.flags.has_kw); + body->param.flags.has_kw = false; + body->param.flags.has_kwrest = true; + body->param.flags.anon_kwrest = true; + body->param.keyword = keyword = ZALLOC_N(struct rb_iseq_param_keyword, 1); + keyword->rest_start = local_index; + pm_insert_local_special(PM_CONSTANT_POW, idPow, local_index++, &index_lookup_table, local_table_for_iseq); + + // Add the anonymous & + body->param.block_start = local_index; + body->param.flags.has_block = true; + pm_insert_local_special(PM_CONSTANT_AND, idAnd, local_index++, &index_lookup_table, local_table_for_iseq); + } + + // Add the ... + pm_insert_local_special(PM_CONSTANT_DOT3, idDot3, local_index++, &index_lookup_table, local_table_for_iseq); + break; + } + default: + rb_bug("node type %s not expected as keyword_rest", pm_node_type(PM_NODE_TYPE(parameters_node->keyword_rest))); + } + } + + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^ + if (parameters_node->block) { + switch (PM_NODE_TYPE(parameters_node->block)) { + case PM_BLOCK_PARAMETER_NODE: { + body->param.block_start = local_index; + body->param.flags.has_block = true; + + iseq_set_use_block(iseq); + + pm_constant_id_t name = ((const pm_block_parameter_node_t *) parameters_node->block)->name; + + if (name) { + if (PM_NODE_FLAG_P(parameters_node->block, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { + ID local = pm_constant_id_lookup(scope_node, name); + local_table_for_iseq->ids[local_index] = local; } else { - static_literal = true; + pm_insert_local_index(name, local_index, &index_lookup_table, local_table_for_iseq, scope_node); } - } else { - PM_COMPILE_NOT_POPPED(element); - if (++new_array_size >= max_new_array_size) FLUSH_CHUNK; - static_literal = false; } + else { + pm_insert_local_special(PM_CONSTANT_AND, idAnd, local_index, &index_lookup_table, local_table_for_iseq); + } + + local_index++; + break; + } + case PM_NO_BLOCK_PARAMETER_NODE: { + body->param.flags.accepts_no_block = true; + break; + } + default: + rb_bug("node type %s not expected as block parameter", pm_node_type(PM_NODE_TYPE(parameters_node->block))); } + } + } - FLUSH_CHUNK; - if (popped) PUSH_INSN(ret, location, pop); + //********END OF STEP 2********** + // The local table is now consistent with expected + // stack layout + + // If there's only one required element in the parameters + // CRuby needs to recognize it as an ambiguous parameter + + //********STEP 3********** + // Goal: fill in the names of the parameters in MultiTargetNodes + // + // Go through requireds again to set the multis + + if (requireds_list && requireds_list->size) { + for (size_t i = 0; i < requireds_list->size; i++) { + // For each MultiTargetNode, we're going to have one + // additional anonymous local not represented in the locals table + // We want to account for this in our table size + const pm_node_t *required = requireds_list->nodes[i]; + + if (PM_NODE_TYPE_P(required, PM_MULTI_TARGET_NODE)) { + local_index = pm_compile_destructured_param_locals((const pm_multi_target_node_t *) required, &index_lookup_table, local_table_for_iseq, scope_node, local_index); + } } + } -#undef FLUSH_CHUNK - return; - } - case PM_ASSOC_NODE: { - // { foo: 1 } - // ^^^^^^ - // - // foo(bar: 1) - // ^^^^^^ - const pm_assoc_node_t *cast = (const pm_assoc_node_t *) node; + // Go through posts again to set the multis + if (posts_list && posts_list->size) { + for (size_t i = 0; i < posts_list->size; i++) { + // For each MultiTargetNode, we're going to have one + // additional anonymous local not represented in the locals table + // We want to account for this in our table size + const pm_node_t *post = posts_list->nodes[i]; - PM_COMPILE(cast->key); - PM_COMPILE(cast->value); + if (PM_NODE_TYPE_P(post, PM_MULTI_TARGET_NODE)) { + local_index = pm_compile_destructured_param_locals((const pm_multi_target_node_t *) post, &index_lookup_table, local_table_for_iseq, scope_node, local_index); + } + } + } - return; + // Set any anonymous locals for the for node + if (PM_NODE_TYPE_P(scope_node->ast_node, PM_FOR_NODE)) { + if (PM_NODE_TYPE_P(((const pm_for_node_t *) scope_node->ast_node)->index, PM_LOCAL_VARIABLE_TARGET_NODE)) { + body->param.lead_num++; + } + else { + body->param.rest_start = local_index; + body->param.flags.has_rest = true; + } + + ID local = rb_make_temporary_id(local_index); + local_table_for_iseq->ids[local_index] = local; + local_index++; + } + + // Fill in any NumberedParameters, if they exist + if (scope_node->parameters && PM_NODE_TYPE_P(scope_node->parameters, PM_NUMBERED_PARAMETERS_NODE)) { + int maximum = ((const pm_numbered_parameters_node_t *) scope_node->parameters)->maximum; + RUBY_ASSERT(0 < maximum && maximum <= 9); + for (int i = 0; i < maximum; i++, local_index++) { + const uint8_t param_name[] = { '_', '1' + i }; + pm_constant_id_t constant_id = pm_parser_constant_find(scope_node->parser, param_name, 2); + RUBY_ASSERT(constant_id && "parser should fill in any gaps in numbered parameters"); + pm_insert_local_index(constant_id, local_index, &index_lookup_table, local_table_for_iseq, scope_node); + } + body->param.lead_num = maximum; + body->param.flags.has_lead = true; + } + + // Fill in the anonymous `it` parameter, if it exists + if (scope_node->parameters && PM_NODE_TYPE_P(scope_node->parameters, PM_IT_PARAMETERS_NODE)) { + body->param.lead_num = 1; + body->param.flags.has_lead = true; + } + + //********END OF STEP 3********** + + //********STEP 4********** + // Goal: fill in the method body locals + // To be explicit, these are the non-parameter locals + // We fill in the block_locals, if they exist + // lambda { |x; y| y } + // ^ + if (block_locals && block_locals->size) { + for (size_t i = 0; i < block_locals->size; i++, local_index++) { + pm_constant_id_t constant_id = ((const pm_block_local_variable_node_t *) block_locals->nodes[i])->name; + pm_insert_local_index(constant_id, local_index, &index_lookup_table, local_table_for_iseq, scope_node); + } + } + + // Fill in any locals we missed + if (scope_node->locals.size) { + for (size_t i = 0; i < scope_node->locals.size; i++) { + pm_constant_id_t constant_id = locals->ids[i]; + if (constant_id) { + int existing; + if (!pm_index_lookup_table_lookup(&index_lookup_table, constant_id, &existing)) { + ID local = pm_constant_id_lookup(scope_node, constant_id); + local_table_for_iseq->ids[local_index] = local; + pm_index_lookup_table_insert(&index_lookup_table, constant_id, local_index); + local_index++; + } + } + } + } + + //********END OF STEP 4********** + + // We set the index_lookup_table on the scope node so we can + // refer to the parameters correctly. + scope_node->index_lookup_table = index_lookup_table; + iseq_calc_param_size(iseq); + + if (ISEQ_BODY(iseq)->param.flags.forwardable) { + // We're treating `...` as a parameter so that frame + // pushing won't clobber it. + ISEQ_BODY(iseq)->param.size += 1; + } + + // FIXME: args? + iseq_set_local_table(iseq, local_table_for_iseq, 0); + iseq_set_parameters_lvar_state(iseq); + + scope_node->local_table_for_iseq_size = local_table_for_iseq->size; + + if (keyword != NULL) { + size_t keyword_start_index = keyword->bits_start - keyword->num; + keyword->table = (ID *)&ISEQ_BODY(iseq)->local_table[keyword_start_index]; + } + + //********STEP 5************ + // Goal: compile anything that needed to be compiled + if (optionals_list && optionals_list->size) { + LABEL **opt_table = (LABEL **) ALLOC_N(VALUE, optionals_list->size + 1); + LABEL *label; + + // TODO: Should we make an api for NEW_LABEL where you can pass + // a pointer to the label it should fill out? We already + // have a list of labels allocated above so it seems wasteful + // to do the copies. + for (size_t i = 0; i < optionals_list->size; i++) { + label = NEW_LABEL(location.line); + opt_table[i] = label; + PUSH_LABEL(ret, label); + pm_node_t *optional_node = optionals_list->nodes[i]; + PM_COMPILE_NOT_POPPED(optional_node); + } + + // Set the last label + label = NEW_LABEL(location.line); + opt_table[optionals_list->size] = label; + PUSH_LABEL(ret, label); + + body->param.opt_table = (const VALUE *) opt_table; + } + + if (keywords_list && keywords_list->size) { + size_t optional_index = 0; + for (size_t i = 0; i < keywords_list->size; i++) { + pm_node_t *keyword_parameter_node = keywords_list->nodes[i]; + pm_constant_id_t name; + + switch (PM_NODE_TYPE(keyword_parameter_node)) { + case PM_OPTIONAL_KEYWORD_PARAMETER_NODE: { + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^^^ + const pm_optional_keyword_parameter_node_t *cast = ((const pm_optional_keyword_parameter_node_t *) keyword_parameter_node); + + pm_node_t *value = cast->value; + name = cast->name; + + if (!PM_NODE_FLAG_P(value, PM_NODE_FLAG_STATIC_LITERAL) || PM_CONTAINER_P(value)) { + LABEL *end_label = NEW_LABEL(location.line); + + pm_local_index_t index = pm_lookup_local_index(iseq, scope_node, name, 0); + int kw_bits_idx = table_size - body->param.keyword->bits_start; + PUSH_INSN2(ret, location, checkkeyword, INT2FIX(kw_bits_idx + VM_ENV_DATA_SIZE - 1), INT2FIX(optional_index)); + PUSH_INSNL(ret, location, branchif, end_label); + PM_COMPILE(value); + PUSH_SETLOCAL(ret, location, index.index, index.level); + PUSH_LABEL(ret, end_label); + } + optional_index++; + break; + } + case PM_REQUIRED_KEYWORD_PARAMETER_NODE: + // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) + // ^^ + break; + default: + rb_bug("Unexpected keyword parameter node type %s", pm_node_type(PM_NODE_TYPE(keyword_parameter_node))); + } + } + } + + if (requireds_list && requireds_list->size) { + for (size_t i = 0; i < requireds_list->size; i++) { + // For each MultiTargetNode, we're going to have one additional + // anonymous local not represented in the locals table. We want + // to account for this in our table size. + const pm_node_t *required = requireds_list->nodes[i]; + + if (PM_NODE_TYPE_P(required, PM_MULTI_TARGET_NODE)) { + PUSH_GETLOCAL(ret, location, table_size - (int)i, 0); + pm_compile_destructured_param_writes(iseq, (const pm_multi_target_node_t *) required, ret, scope_node); + } + } + } + + if (posts_list && posts_list->size) { + for (size_t i = 0; i < posts_list->size; i++) { + // For each MultiTargetNode, we're going to have one additional + // anonymous local not represented in the locals table. We want + // to account for this in our table size. + const pm_node_t *post = posts_list->nodes[i]; + + if (PM_NODE_TYPE_P(post, PM_MULTI_TARGET_NODE)) { + PUSH_GETLOCAL(ret, location, table_size - body->param.post_start - (int) i, 0); + pm_compile_destructured_param_writes(iseq, (const pm_multi_target_node_t *) post, ret, scope_node); + } + } + } + + switch (body->type) { + case ISEQ_TYPE_PLAIN: { + RUBY_ASSERT(PM_NODE_TYPE_P(scope_node->ast_node, PM_INTERPOLATED_REGULAR_EXPRESSION_NODE)); + + const pm_interpolated_regular_expression_node_t *cast = (const pm_interpolated_regular_expression_node_t *) scope_node->ast_node; + pm_compile_regexp_dynamic(iseq, (const pm_node_t *) cast, &cast->parts, &location, ret, popped, scope_node); + + break; } - case PM_ASSOC_SPLAT_NODE: { - // { **foo } - // ^^^^^ - // - // def foo(**); bar(**); end - // ^^ - const pm_assoc_splat_node_t *cast = (const pm_assoc_splat_node_t *) node; + case ISEQ_TYPE_BLOCK: { + LABEL *start = ISEQ_COMPILE_DATA(iseq)->start_label = NEW_LABEL(0); + LABEL *end = ISEQ_COMPILE_DATA(iseq)->end_label = NEW_LABEL(0); + const pm_node_location_t block_location = { .line = body->location.first_lineno, .node_id = scope_node->ast_node->node_id }; - if (cast->value != NULL) { - PM_COMPILE(cast->value); + start->rescued = LABEL_RESCUE_BEG; + end->rescued = LABEL_RESCUE_END; + + // For nodes automatically assign the iteration variable to whatever + // index variable. We need to handle that write here because it has + // to happen in the context of the block. Note that this happens + // before the B_CALL tracepoint event. + if (PM_NODE_TYPE_P(scope_node->ast_node, PM_FOR_NODE)) { + pm_compile_for_node_index(iseq, ((const pm_for_node_t *) scope_node->ast_node)->index, ret, scope_node); } - else if (!popped) { - pm_local_index_t index = pm_lookup_local_index(iseq, scope_node, PM_CONSTANT_POW, 0); - PUSH_GETLOCAL(ret, location, index.index, index.level); + + PUSH_TRACE(ret, RUBY_EVENT_B_CALL); + PUSH_INSN(ret, block_location, nop); + PUSH_LABEL(ret, start); + + if (scope_node->body != NULL) { + switch (PM_NODE_TYPE(scope_node->ast_node)) { + case PM_POST_EXECUTION_NODE: { + const pm_post_execution_node_t *cast = (const pm_post_execution_node_t *) scope_node->ast_node; + PUSH_INSN1(ret, block_location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); + + // We create another ScopeNode from the statements within the PostExecutionNode + pm_scope_node_t next_scope_node; + pm_scope_node_init((const pm_node_t *) cast->statements, &next_scope_node, scope_node); + + const rb_iseq_t *block = NEW_CHILD_ISEQ(&next_scope_node, make_name_for_block(body->parent_iseq), ISEQ_TYPE_BLOCK, location.line); + pm_scope_node_destroy(&next_scope_node); + + PUSH_CALL_WITH_BLOCK(ret, block_location, id_core_set_postexe, INT2FIX(0), block); + break; + } + case PM_INTERPOLATED_REGULAR_EXPRESSION_NODE: { + const pm_interpolated_regular_expression_node_t *cast = (const pm_interpolated_regular_expression_node_t *) scope_node->ast_node; + pm_compile_regexp_dynamic(iseq, (const pm_node_t *) cast, &cast->parts, &location, ret, popped, scope_node); + break; + } + default: + pm_compile_node(iseq, scope_node->body, ret, popped, scope_node); + break; + } + } + else { + PUSH_INSN(ret, block_location, putnil); } - return; + PUSH_LABEL(ret, end); + PUSH_TRACE(ret, RUBY_EVENT_B_RETURN); + ISEQ_COMPILE_DATA(iseq)->last_line = body->location.code_location.end_pos.lineno; + + /* wide range catch handler must put at last */ + PUSH_CATCH_ENTRY(CATCH_TYPE_REDO, start, end, NULL, start); + PUSH_CATCH_ENTRY(CATCH_TYPE_NEXT, start, end, NULL, end); + break; } - case PM_BACK_REFERENCE_READ_NODE: { - // $+ - // ^^ - if (!popped) { - // Since a back reference is `$<char>`, ruby represents the ID as the - // an rb_intern on the value after the `$`. - char *char_ptr = (char *)(node->location.start) + 1; - ID backref_val = INT2FIX(rb_intern2(char_ptr, 1)) << 1 | 1; - PUSH_INSN2(ret, location, getspecial, INT2FIX(1), backref_val); + case ISEQ_TYPE_ENSURE: { + const pm_node_location_t statements_location = (scope_node->body != NULL ? PM_NODE_START_LOCATION(scope_node->body) : location); + iseq_set_exception_local_table(iseq); + + if (scope_node->body != NULL) { + PM_COMPILE_POPPED((const pm_node_t *) scope_node->body); } + + PUSH_GETLOCAL(ret, statements_location, 1, 0); + PUSH_INSN1(ret, statements_location, throw, INT2FIX(0)); return; } - case PM_BEGIN_NODE: { - // begin end - // ^^^^^^^^^ - const pm_begin_node_t *cast = (const pm_begin_node_t *) node; + case ISEQ_TYPE_METHOD: { + ISEQ_COMPILE_DATA(iseq)->root_node = (const void *) scope_node->body; + PUSH_TRACE(ret, RUBY_EVENT_CALL); - if (cast->ensure_clause) { - // Compiling the ensure clause will compile the rescue clause (if - // there is one), which will compile the begin statements. - pm_compile_ensure(iseq, cast, &location, ret, popped, scope_node); + if (scope_node->body) { + PM_COMPILE((const pm_node_t *) scope_node->body); } - else if (cast->rescue_clause) { - // Compiling rescue will compile begin statements (if applicable). - pm_compile_rescue(iseq, cast, &location, ret, popped, scope_node); + else { + PUSH_INSN(ret, location, putnil); + } + + ISEQ_COMPILE_DATA(iseq)->root_node = (const void *) scope_node->body; + PUSH_TRACE(ret, RUBY_EVENT_RETURN); + + ISEQ_COMPILE_DATA(iseq)->last_line = body->location.code_location.end_pos.lineno; + break; + } + case ISEQ_TYPE_RESCUE: { + iseq_set_exception_local_table(iseq); + if (PM_NODE_TYPE_P(scope_node->ast_node, PM_RESCUE_MODIFIER_NODE)) { + LABEL *lab = NEW_LABEL(location.line); + LABEL *rescue_end = NEW_LABEL(location.line); + PUSH_GETLOCAL(ret, location, LVAR_ERRINFO, 0); + PUSH_INSN1(ret, location, putobject, rb_eStandardError); + PUSH_INSN1(ret, location, checkmatch, INT2FIX(VM_CHECKMATCH_TYPE_RESCUE)); + PUSH_INSNL(ret, location, branchif, lab); + PUSH_INSNL(ret, location, jump, rescue_end); + PUSH_LABEL(ret, lab); + PUSH_TRACE(ret, RUBY_EVENT_RESCUE); + PM_COMPILE((const pm_node_t *) scope_node->body); + PUSH_INSN(ret, location, leave); + PUSH_LABEL(ret, rescue_end); + PUSH_GETLOCAL(ret, location, LVAR_ERRINFO, 0); } else { - // If there is neither ensure or rescue, the just compile the - // statements. - if (cast->statements != NULL) { - PM_COMPILE((const pm_node_t *) cast->statements); - } - else if (!popped) { - PUSH_SYNTHETIC_PUTNIL(ret, iseq); - } + PM_COMPILE((const pm_node_t *) scope_node->ast_node); } + PUSH_INSN1(ret, location, throw, INT2FIX(0)); + return; } - case PM_BLOCK_ARGUMENT_NODE: { - // foo(&bar) - // ^^^^ - const pm_block_argument_node_t *cast = (const pm_block_argument_node_t *) node; - - if (cast->expression != NULL) { - PM_COMPILE(cast->expression); + default: + if (scope_node->body) { + PM_COMPILE((const pm_node_t *) scope_node->body); } else { - // If there's no expression, this must be block forwarding. - pm_local_index_t local_index = pm_lookup_local_index(iseq, scope_node, PM_CONSTANT_AND, 0); - PUSH_INSN2(ret, location, getblockparamproxy, INT2FIX(local_index.index + VM_ENV_DATA_SIZE - 1), INT2FIX(local_index.level)); + PUSH_INSN(ret, location, putnil); + } + break; + } + + if (PM_NODE_TYPE_P(scope_node->ast_node, PM_CLASS_NODE) || PM_NODE_TYPE_P(scope_node->ast_node, PM_MODULE_NODE)) { + const pm_node_location_t end_location = PM_NODE_END_LOCATION(scope_node->ast_node); + PUSH_TRACE(ret, RUBY_EVENT_END); + ISEQ_COMPILE_DATA(iseq)->last_line = end_location.line; + } + + if (!PM_NODE_TYPE_P(scope_node->ast_node, PM_ENSURE_NODE)) { + const pm_node_location_t location = { .line = ISEQ_COMPILE_DATA(iseq)->last_line, .node_id = scope_node->ast_node->node_id }; + PUSH_INSN(ret, location, leave); + } +} + +static inline void +pm_compile_alias_global_variable_node(rb_iseq_t *iseq, const pm_alias_global_variable_node_t *node, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + // alias $foo $bar + // ^^^^^^^^^^^^^^^ + PUSH_INSN1(ret, *location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); + + { + const pm_location_t *name_loc = &node->new_name->location; + VALUE operand = ID2SYM(rb_intern3((const char *) (pm_parser_start(scope_node->parser) + name_loc->start), name_loc->length, scope_node->encoding)); + PUSH_INSN1(ret, *location, putobject, operand); + } + + { + const pm_location_t *name_loc = &node->old_name->location; + VALUE operand = ID2SYM(rb_intern3((const char *) (pm_parser_start(scope_node->parser) + name_loc->start), name_loc->length, scope_node->encoding)); + PUSH_INSN1(ret, *location, putobject, operand); + } + + PUSH_SEND(ret, *location, id_core_set_variable_alias, INT2FIX(2)); + if (popped) PUSH_INSN(ret, *location, pop); +} + +static inline void +pm_compile_alias_method_node(rb_iseq_t *iseq, const pm_alias_method_node_t *node, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + PUSH_INSN1(ret, *location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); + PUSH_INSN1(ret, *location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_CBASE)); + PM_COMPILE_NOT_POPPED(node->new_name); + PM_COMPILE_NOT_POPPED(node->old_name); + + PUSH_SEND(ret, *location, id_core_set_method_alias, INT2FIX(3)); + if (popped) PUSH_INSN(ret, *location, pop); +} + +static inline void +pm_compile_and_node(rb_iseq_t *iseq, const pm_and_node_t *node, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + LABEL *end_label = NEW_LABEL(location->line); + + PM_COMPILE_NOT_POPPED(node->left); + if (!popped) PUSH_INSN(ret, *location, dup); + PUSH_INSNL(ret, *location, branchunless, end_label); + + if (!popped) PUSH_INSN(ret, *location, pop); + PM_COMPILE(node->right); + PUSH_LABEL(ret, end_label); +} + +static inline void +pm_compile_array_node(rb_iseq_t *iseq, const pm_node_t *node, const pm_node_list_t *elements, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + // If every node in the array is static, then we can compile the entire + // array now instead of later. + if (PM_NODE_FLAG_P(node, PM_NODE_FLAG_STATIC_LITERAL)) { + // We're only going to compile this node if it's not popped. If it + // is popped, then we know we don't need to do anything since it's + // statically known. + if (!popped) { + if (elements->size) { + VALUE value = pm_static_literal_value(iseq, node, scope_node); + RB_OBJ_SET_FROZEN_SHAREABLE(value); + PUSH_INSN1(ret, *location, duparray, value); + } + else { + PUSH_INSN1(ret, *location, newarray, INT2FIX(0)); + } } return; - } - case PM_BREAK_NODE: { - // break - // ^^^^^ - // - // break foo - // ^^^^^^^^^ - const pm_break_node_t *cast = (const pm_break_node_t *) node; - unsigned long throw_flag = 0; + } + + // Here since we know there are possible side-effects inside the + // array contents, we're going to build it entirely at runtime. + // We'll do this by pushing all of the elements onto the stack and + // then combining them with newarray. + // + // If this array is popped, then this serves only to ensure we enact + // all side-effects (like method calls) that are contained within + // the array contents. + // + // We treat all sequences of non-splat elements as their + // own arrays, followed by a newarray, and then continually + // concat the arrays with the SplatNode nodes. + const int max_new_array_size = 0x100; + const unsigned int min_tmp_array_size = 0x40; + + int new_array_size = 0; + bool first_chunk = true; - if (ISEQ_COMPILE_DATA(iseq)->redo_label != 0 && can_add_ensure_iseq(iseq)) { - /* while/until */ - LABEL *splabel = NEW_LABEL(0); - PUSH_LABEL(ret, splabel); - PUSH_ADJUST(ret, location, ISEQ_COMPILE_DATA(iseq)->redo_label); + // This is an optimization wherein we keep track of whether or not + // the previous element was a static literal. If it was, then we do + // not attempt to check if we have a subarray that can be optimized. + // If it was not, then we do check. + bool static_literal = false; + + // Either create a new array, or push to the existing array. +#define FLUSH_CHUNK \ + if (new_array_size) { \ + if (first_chunk) PUSH_INSN1(ret, *location, newarray, INT2FIX(new_array_size)); \ + else PUSH_INSN1(ret, *location, pushtoarray, INT2FIX(new_array_size)); \ + first_chunk = false; \ + new_array_size = 0; \ + } - if (cast->arguments != NULL) { - PM_COMPILE_NOT_POPPED((const pm_node_t *) cast->arguments); + for (size_t index = 0; index < elements->size; index++) { + const pm_node_t *element = elements->nodes[index]; + + if (PM_NODE_TYPE_P(element, PM_SPLAT_NODE)) { + FLUSH_CHUNK; + + const pm_splat_node_t *splat_element = (const pm_splat_node_t *) element; + if (splat_element->expression) { + PM_COMPILE_NOT_POPPED(splat_element->expression); } else { - PUSH_INSN(ret, location, putnil); + pm_local_index_t index = pm_lookup_local_index(iseq, scope_node, PM_CONSTANT_MULT, 0); + PUSH_GETLOCAL(ret, *location, index.index, index.level); } - pm_add_ensure_iseq(ret, iseq, 0, scope_node); - PUSH_INSNL(ret, location, jump, ISEQ_COMPILE_DATA(iseq)->end_label); - PUSH_ADJUST_RESTORE(ret, splabel); - if (!popped) PUSH_INSN(ret, location, putnil); + if (first_chunk) { + // If this is the first element of the array then we + // need to splatarray the elements into the list. + PUSH_INSN1(ret, *location, splatarray, Qtrue); + first_chunk = false; + } + else { + PUSH_INSN(ret, *location, concattoarray); + } + + static_literal = false; } - else { - const rb_iseq_t *ip = iseq; + else if (PM_NODE_TYPE_P(element, PM_KEYWORD_HASH_NODE)) { + if (new_array_size == 0 && first_chunk) { + PUSH_INSN1(ret, *location, newarray, INT2FIX(0)); + first_chunk = false; + } + else { + FLUSH_CHUNK; + } - while (ip) { - if (!ISEQ_COMPILE_DATA(ip)) { - ip = 0; - break; - } + // If we get here, then this is the last element of the + // array/arguments, because it cannot be followed by + // anything else without a syntax error. This looks like: + // + // [foo, bar, baz: qux] + // ^^^^^^^^ + // + // [foo, bar, **baz] + // ^^^^^ + // + const pm_keyword_hash_node_t *keyword_hash = (const pm_keyword_hash_node_t *) element; + pm_compile_hash_elements(iseq, element, &keyword_hash->elements, 0, Qundef, false, ret, scope_node); - if (ISEQ_COMPILE_DATA(ip)->redo_label != 0) { - throw_flag = VM_THROW_NO_ESCAPE_FLAG; - } - else if (ISEQ_BODY(ip)->type == ISEQ_TYPE_BLOCK) { - throw_flag = 0; - } - else if (ISEQ_BODY(ip)->type == ISEQ_TYPE_EVAL) { - COMPILE_ERROR(iseq, location.line, "Can't escape from eval with break"); - return; - } - else { - ip = ISEQ_BODY(ip)->parent_iseq; - continue; - } + // This boolean controls the manner in which we push the + // hash onto the array. If it's all keyword splats, then we + // can use the very specialized pushtoarraykwsplat + // instruction to check if it's empty before we push it. + size_t splats = 0; + while (splats < keyword_hash->elements.size && PM_NODE_TYPE_P(keyword_hash->elements.nodes[splats], PM_ASSOC_SPLAT_NODE)) splats++; - /* escape from block */ - if (cast->arguments != NULL) { - PM_COMPILE_NOT_POPPED((const pm_node_t *) cast->arguments); + if (keyword_hash->elements.size == splats) { + PUSH_INSN(ret, *location, pushtoarraykwsplat); + } + else { + new_array_size++; + } + } + else if ( + PM_NODE_FLAG_P(element, PM_NODE_FLAG_STATIC_LITERAL) && + !PM_CONTAINER_P(element) && + !static_literal && + ((index + min_tmp_array_size) < elements->size) + ) { + // If we have a static literal, then there's the potential + // to group a bunch of them together with a literal array + // and then concat them together. + size_t right_index = index + 1; + while ( + right_index < elements->size && + PM_NODE_FLAG_P(elements->nodes[right_index], PM_NODE_FLAG_STATIC_LITERAL) && + !PM_CONTAINER_P(elements->nodes[right_index]) + ) right_index++; + + size_t tmp_array_size = right_index - index; + if (tmp_array_size >= min_tmp_array_size) { + VALUE tmp_array = rb_ary_hidden_new(tmp_array_size); + + // Create the temporary array. + for (; tmp_array_size; tmp_array_size--) + rb_ary_push(tmp_array, pm_static_literal_value(iseq, elements->nodes[index++], scope_node)); + + index--; // about to be incremented by for loop + RB_OBJ_SET_FROZEN_SHAREABLE(tmp_array); + + // Emit the optimized code. + FLUSH_CHUNK; + if (first_chunk) { + PUSH_INSN1(ret, *location, duparray, tmp_array); + first_chunk = false; } else { - PUSH_INSN(ret, location, putnil); + PUSH_INSN1(ret, *location, putobject, tmp_array); + PUSH_INSN(ret, *location, concattoarray); } + } + else { + PM_COMPILE_NOT_POPPED(element); + if (++new_array_size >= max_new_array_size) FLUSH_CHUNK; + static_literal = true; + } + } else { + PM_COMPILE_NOT_POPPED(element); + if (++new_array_size >= max_new_array_size) FLUSH_CHUNK; + static_literal = false; + } + } + + FLUSH_CHUNK; + if (popped) PUSH_INSN(ret, *location, pop); + +#undef FLUSH_CHUNK +} + +static inline void +pm_compile_break_node(rb_iseq_t *iseq, const pm_break_node_t *node, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + unsigned long throw_flag = 0; + + if (ISEQ_COMPILE_DATA(iseq)->redo_label != 0 && can_add_ensure_iseq(iseq)) { + /* while/until */ + LABEL *splabel = NEW_LABEL(0); + PUSH_LABEL(ret, splabel); + PUSH_ADJUST(ret, *location, ISEQ_COMPILE_DATA(iseq)->redo_label); + + if (node->arguments != NULL) { + PM_COMPILE_NOT_POPPED((const pm_node_t *) node->arguments); + } + else { + PUSH_INSN(ret, *location, putnil); + } + + pm_add_ensure_iseq(ret, iseq, 0, scope_node); + PUSH_INSNL(ret, *location, jump, ISEQ_COMPILE_DATA(iseq)->end_label); + PUSH_ADJUST_RESTORE(ret, splabel); + if (!popped) PUSH_INSN(ret, *location, putnil); + } + else { + const rb_iseq_t *ip = iseq; - PUSH_INSN1(ret, location, throw, INT2FIX(throw_flag | TAG_BREAK)); - if (popped) PUSH_INSN(ret, location, pop); + while (ip) { + if (!ISEQ_COMPILE_DATA(ip)) { + ip = 0; + break; + } + if (ISEQ_COMPILE_DATA(ip)->redo_label != 0) { + throw_flag = VM_THROW_NO_ESCAPE_FLAG; + } + else if (ISEQ_BODY(ip)->type == ISEQ_TYPE_BLOCK) { + throw_flag = 0; + } + else if (ISEQ_BODY(ip)->type == ISEQ_TYPE_EVAL) { + COMPILE_ERROR(iseq, location->line, "Invalid break"); return; } + else { + ip = ISEQ_BODY(ip)->parent_iseq; + continue; + } - COMPILE_ERROR(iseq, location.line, "Invalid break"); + /* escape from block */ + if (node->arguments != NULL) { + PM_COMPILE_NOT_POPPED((const pm_node_t *) node->arguments); + } + else { + PUSH_INSN(ret, *location, putnil); + } + + PUSH_INSN1(ret, *location, throw, INT2FIX(throw_flag | TAG_BREAK)); + if (popped) PUSH_INSN(ret, *location, pop); + + return; } - return; - } - case PM_CALL_NODE: { - // foo - // ^^^ - // - // foo.bar - // ^^^^^^^ - // - // foo.bar() {} - // ^^^^^^^^^^^^ - const pm_call_node_t *cast = (const pm_call_node_t *) node; - ID method_id = pm_constant_id_lookup(scope_node, cast->name); - const pm_location_t *message_loc = &cast->message_loc; - if (message_loc->start == NULL) message_loc = &cast->base.location; + COMPILE_ERROR(iseq, location->line, "Invalid break"); + } +} + +static inline void +pm_compile_call_node(rb_iseq_t *iseq, const pm_call_node_t *node, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + ID method_id = pm_constant_id_lookup(scope_node, node->name); + + const pm_location_t *message_loc = &node->message_loc; + if (message_loc->length == 0) message_loc = &node->base.location; - const pm_node_location_t location = PM_LOCATION_START_LOCATION(scope_node->parser, message_loc, cast->base.node_id); - const char *builtin_func; + const pm_node_location_t location = PM_LOCATION_START_LOCATION(message_loc, node->base.node_id); + const char *builtin_func; - if (UNLIKELY(iseq_has_builtin_function_table(iseq)) && (builtin_func = pm_iseq_builtin_function_name(scope_node, cast->receiver, method_id)) != NULL) { - pm_compile_builtin_function_call(iseq, ret, scope_node, cast, &location, popped, ISEQ_COMPILE_DATA(iseq)->current_block, builtin_func); + if (UNLIKELY(iseq_has_builtin_function_table(iseq)) && (builtin_func = pm_iseq_builtin_function_name(scope_node, node->receiver, method_id)) != NULL) { + pm_compile_builtin_function_call(iseq, ret, scope_node, node, &location, popped, ISEQ_COMPILE_DATA(iseq)->current_block, builtin_func); + return; + } + + LABEL *start = NEW_LABEL(location.line); + if (node->block) PUSH_LABEL(ret, start); + + switch (method_id) { + case idUMinus: { + if (pm_opt_str_freeze_p(iseq, node)) { + VALUE value = parse_static_literal_string(iseq, scope_node, node->receiver, &((const pm_string_node_t * ) node->receiver)->unescaped); + const struct rb_callinfo *callinfo = new_callinfo(iseq, idUMinus, 0, 0, NULL, FALSE); + PUSH_INSN2(ret, location, opt_str_uminus, value, callinfo); + if (popped) PUSH_INSN(ret, location, pop); + return; + } + break; + } + case idFreeze: { + if (pm_opt_str_freeze_p(iseq, node)) { + VALUE value = parse_static_literal_string(iseq, scope_node, node->receiver, &((const pm_string_node_t * ) node->receiver)->unescaped); + const struct rb_callinfo *callinfo = new_callinfo(iseq, idFreeze, 0, 0, NULL, FALSE); + PUSH_INSN2(ret, location, opt_str_freeze, value, callinfo); + if (popped) PUSH_INSN(ret, location, pop); return; } + break; + } + } - LABEL *start = NEW_LABEL(location.line); - if (cast->block) PUSH_LABEL(ret, start); + if (PM_NODE_FLAG_P(node, PM_CALL_NODE_FLAGS_ATTRIBUTE_WRITE) && !popped) { + PUSH_INSN(ret, location, putnil); + } - switch (method_id) { - case idUMinus: { - if (pm_opt_str_freeze_p(iseq, cast)) { - VALUE value = parse_static_literal_string(iseq, scope_node, cast->receiver, &((const pm_string_node_t * ) cast->receiver)->unescaped); - PUSH_INSN2(ret, location, opt_str_uminus, value, new_callinfo(iseq, idUMinus, 0, 0, NULL, FALSE)); - return; + if (node->receiver == NULL) { + PUSH_INSN(ret, location, putself); + } + else { + if (method_id == idCall && PM_NODE_TYPE_P(node->receiver, PM_LOCAL_VARIABLE_READ_NODE)) { + const pm_local_variable_read_node_t *read_node_cast = (const pm_local_variable_read_node_t *) node->receiver; + uint32_t node_id = node->receiver->node_id; + int idx, level; + + if (iseq_block_param_id_p(iseq, pm_constant_id_lookup(scope_node, read_node_cast->name), &idx, &level)) { + ADD_ELEM(ret, (LINK_ELEMENT *) new_insn_body(iseq, location.line, node_id, BIN(getblockparamproxy), 2, INT2FIX((idx) + VM_ENV_DATA_SIZE - 1), INT2FIX(level))); } - break; - } - case idFreeze: { - if (pm_opt_str_freeze_p(iseq, cast)) { - VALUE value = parse_static_literal_string(iseq, scope_node, cast->receiver, &((const pm_string_node_t * ) cast->receiver)->unescaped); - PUSH_INSN2(ret, location, opt_str_freeze, value, new_callinfo(iseq, idFreeze, 0, 0, NULL, FALSE)); - return; + else { + PM_COMPILE_NOT_POPPED(node->receiver); } - break; - } - case idAREF: { - if (pm_opt_aref_with_p(iseq, cast)) { - const pm_string_node_t *string = (const pm_string_node_t *) ((const pm_arguments_node_t *) cast->arguments)->arguments.nodes[0]; - VALUE value = parse_static_literal_string(iseq, scope_node, (const pm_node_t *) string, &string->unescaped); + } + else { + PM_COMPILE_NOT_POPPED(node->receiver); + } + } - PM_COMPILE_NOT_POPPED(cast->receiver); - PUSH_INSN2(ret, location, opt_aref_with, value, new_callinfo(iseq, idAREF, 1, 0, NULL, FALSE)); + pm_compile_call(iseq, node, ret, popped, scope_node, method_id, start); + return; +} - if (popped) { - PUSH_INSN(ret, location, pop); - } +static inline void +pm_compile_call_operator_write_node(rb_iseq_t *iseq, const pm_call_operator_write_node_t *node, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + int flag = 0; - return; - } - break; - } - case idASET: { - if (pm_opt_aset_with_p(iseq, cast)) { - const pm_string_node_t *string = (const pm_string_node_t *) ((const pm_arguments_node_t *) cast->arguments)->arguments.nodes[0]; - VALUE value = parse_static_literal_string(iseq, scope_node, (const pm_node_t *) string, &string->unescaped); + if (PM_NODE_FLAG_P(node, PM_CALL_NODE_FLAGS_IGNORE_VISIBILITY)) { + flag = VM_CALL_FCALL; + } - PM_COMPILE_NOT_POPPED(cast->receiver); - PM_COMPILE_NOT_POPPED(((const pm_arguments_node_t *) cast->arguments)->arguments.nodes[1]); + PM_COMPILE_NOT_POPPED(node->receiver); - if (!popped) { - PUSH_INSN(ret, location, swap); - PUSH_INSN1(ret, location, topn, INT2FIX(1)); - } + LABEL *safe_label = NULL; + if (PM_NODE_FLAG_P(node, PM_CALL_NODE_FLAGS_SAFE_NAVIGATION)) { + safe_label = NEW_LABEL(location->line); + PUSH_INSN(ret, *location, dup); + PUSH_INSNL(ret, *location, branchnil, safe_label); + } - PUSH_INSN2(ret, location, opt_aset_with, value, new_callinfo(iseq, idASET, 2, 0, NULL, FALSE)); - PUSH_INSN(ret, location, pop); - return; - } - break; - } - } + PUSH_INSN(ret, *location, dup); - if (PM_NODE_FLAG_P(cast, PM_CALL_NODE_FLAGS_ATTRIBUTE_WRITE) && !popped) { - PUSH_INSN(ret, location, putnil); - } + ID id_read_name = pm_constant_id_lookup(scope_node, node->read_name); + PUSH_SEND_WITH_FLAG(ret, *location, id_read_name, INT2FIX(0), INT2FIX(flag)); - if (cast->receiver == NULL) { - PUSH_INSN(ret, location, putself); - } - else { - if (method_id == idCall && PM_NODE_TYPE_P(cast->receiver, PM_LOCAL_VARIABLE_READ_NODE)) { - const pm_local_variable_read_node_t *read_node_cast = (const pm_local_variable_read_node_t *) cast->receiver; - uint32_t node_id = cast->receiver->node_id; - int idx, level; + PM_COMPILE_NOT_POPPED(node->value); + ID id_operator = pm_constant_id_lookup(scope_node, node->binary_operator); + PUSH_SEND(ret, *location, id_operator, INT2FIX(1)); - if (iseq_block_param_id_p(iseq, pm_constant_id_lookup(scope_node, read_node_cast->name), &idx, &level)) { - ADD_ELEM(ret, (LINK_ELEMENT *) new_insn_body(iseq, location.line, node_id, BIN(getblockparamproxy), 2, INT2FIX((idx) + VM_ENV_DATA_SIZE - 1), INT2FIX(level))); - } - else { - PM_COMPILE_NOT_POPPED(cast->receiver); - } - } - else { - PM_COMPILE_NOT_POPPED(cast->receiver); - } + if (!popped) { + PUSH_INSN(ret, *location, swap); + PUSH_INSN1(ret, *location, topn, INT2FIX(1)); + } + + ID id_write_name = pm_constant_id_lookup(scope_node, node->write_name); + PUSH_SEND_WITH_FLAG(ret, *location, id_write_name, INT2FIX(1), INT2FIX(flag)); + + if (safe_label != NULL && popped) PUSH_LABEL(ret, safe_label); + PUSH_INSN(ret, *location, pop); + if (safe_label != NULL && !popped) PUSH_LABEL(ret, safe_label); +} + +/** + * When we're compiling a case node, it's possible that we can speed it up using + * a dispatch hash, which will allow us to jump directly to the correct when + * clause body based on a hash lookup of the value. This can only happen when + * the conditions are literals that can be compiled into a hash key. + * + * This function accepts a dispatch hash and the condition of a when clause. It + * is responsible for compiling the condition into a hash key and then adding it + * to the dispatch hash. + * + * If the value can be successfully compiled into the hash, then this function + * returns the dispatch hash with the new key added. If the value cannot be + * compiled into the hash, then this function returns Qundef. In the case of + * Qundef, this function is signaling that the caller should abandon the + * optimization entirely. + */ +static VALUE +pm_compile_case_node_dispatch(rb_iseq_t *iseq, VALUE dispatch, const pm_node_t *node, LABEL *label, pm_scope_node_t *scope_node) +{ + VALUE key = Qundef; + switch (PM_NODE_TYPE(node)) { + case PM_FLOAT_NODE: { + key = pm_static_literal_value(iseq, node, scope_node); + double intptr; + + if (modf(RFLOAT_VALUE(key), &intptr) == 0.0) { + key = (FIXABLE(intptr) ? LONG2FIX((long) intptr) : rb_dbl2big(intptr)); } - pm_compile_call(iseq, cast, ret, popped, scope_node, method_id, start); - return; - } - case PM_CALL_AND_WRITE_NODE: { - // foo.bar &&= baz - // ^^^^^^^^^^^^^^^ - const pm_call_and_write_node_t *cast = (const pm_call_and_write_node_t *) node; - pm_compile_call_and_or_write_node(iseq, true, cast->receiver, cast->value, cast->write_name, cast->read_name, PM_NODE_FLAG_P(cast, PM_CALL_NODE_FLAGS_SAFE_NAVIGATION), &location, ret, popped, scope_node); - return; + break; } - case PM_CALL_OR_WRITE_NODE: { - // foo.bar ||= baz - // ^^^^^^^^^^^^^^^ - const pm_call_or_write_node_t *cast = (const pm_call_or_write_node_t *) node; - pm_compile_call_and_or_write_node(iseq, false, cast->receiver, cast->value, cast->write_name, cast->read_name, PM_NODE_FLAG_P(cast, PM_CALL_NODE_FLAGS_SAFE_NAVIGATION), &location, ret, popped, scope_node); - return; + case PM_FALSE_NODE: + case PM_INTEGER_NODE: + case PM_NIL_NODE: + case PM_SOURCE_FILE_NODE: + case PM_SOURCE_LINE_NODE: + case PM_SYMBOL_NODE: + case PM_TRUE_NODE: + key = pm_static_literal_value(iseq, node, scope_node); + break; + case PM_STRING_NODE: { + const pm_string_node_t *cast = (const pm_string_node_t *) node; + key = parse_static_literal_string(iseq, scope_node, node, &cast->unescaped); + break; } - case PM_CALL_OPERATOR_WRITE_NODE: { - // foo.bar += baz - // ^^^^^^^^^^^^^^^ - // - // Call operator writes occur when you have a call node on the left-hand - // side of a write operator that is not `=`. As an example, - // `foo.bar *= 1`. This breaks down to caching the receiver on the - // stack and then performing three method calls, one to read the value, - // one to compute the result, and one to write the result back to the - // receiver. - const pm_call_operator_write_node_t *cast = (const pm_call_operator_write_node_t *) node; - int flag = 0; + default: + return Qundef; + } - if (PM_NODE_FLAG_P(cast, PM_CALL_NODE_FLAGS_IGNORE_VISIBILITY)) { - flag = VM_CALL_FCALL; - } + cdhash_aset_if_missing(dispatch, key, (VALUE)label); + return dispatch; +} - PM_COMPILE_NOT_POPPED(cast->receiver); +/** + * Compile a case node, representing a case statement with when clauses. + */ +static inline void +pm_compile_case_node(rb_iseq_t *iseq, const pm_case_node_t *cast, const pm_node_location_t *node_location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + const pm_node_location_t location = *node_location; + const pm_node_list_t *conditions = &cast->conditions; - LABEL *safe_label = NULL; - if (PM_NODE_FLAG_P(cast, PM_CALL_NODE_FLAGS_SAFE_NAVIGATION)) { - safe_label = NEW_LABEL(location.line); - PUSH_INSN(ret, location, dup); - PUSH_INSNL(ret, location, branchnil, safe_label); - } + // This is the anchor that we will compile the conditions of the various + // `when` nodes into. If a match is found, they will need to jump into + // the body_seq anchor to the correct spot. + DECL_ANCHOR(cond_seq); - PUSH_INSN(ret, location, dup); + // This is the anchor that we will compile the bodies of the various + // `when` nodes into. We'll make sure that the clauses that are compiled + // jump into the correct spots within this anchor. + DECL_ANCHOR(body_seq); - ID id_read_name = pm_constant_id_lookup(scope_node, cast->read_name); - PUSH_SEND_WITH_FLAG(ret, location, id_read_name, INT2FIX(0), INT2FIX(flag)); + // This is the label where all of the when clauses will jump to if they + // have matched and are done executing their bodies. + LABEL *end_label = NEW_LABEL(location.line); - PM_COMPILE_NOT_POPPED(cast->value); - ID id_operator = pm_constant_id_lookup(scope_node, cast->binary_operator); - PUSH_SEND(ret, location, id_operator, INT2FIX(1)); + // If we have a predicate on this case statement, then it's going to + // compare all of the various when clauses to the predicate. If we + // don't, then it's basically an if-elsif-else chain. + if (cast->predicate == NULL) { + // Establish branch coverage for the case node. + VALUE branches = Qfalse; + rb_code_location_t case_location = { 0 }; + int branch_id = 0; - if (!popped) { - PUSH_INSN(ret, location, swap); - PUSH_INSN1(ret, location, topn, INT2FIX(1)); + if (PM_BRANCH_COVERAGE_P(iseq)) { + case_location = pm_code_location(scope_node, (const pm_node_t *) cast); + branches = decl_branch_base(iseq, PTR2NUM(cast), &case_location, "case"); } - ID id_write_name = pm_constant_id_lookup(scope_node, cast->write_name); - PUSH_SEND_WITH_FLAG(ret, location, id_write_name, INT2FIX(1), INT2FIX(flag)); + // Loop through each clauses in the case node and compile each of + // the conditions within them into cond_seq. If they match, they + // should jump into their respective bodies in body_seq. + for (size_t clause_index = 0; clause_index < conditions->size; clause_index++) { + const pm_when_node_t *clause = (const pm_when_node_t *) conditions->nodes[clause_index]; + const pm_node_list_t *conditions = &clause->conditions; - if (safe_label != NULL && popped) PUSH_LABEL(ret, safe_label); - PUSH_INSN(ret, location, pop); - if (safe_label != NULL && !popped) PUSH_LABEL(ret, safe_label); + int clause_lineno = pm_node_line_number_cached((const pm_node_t *) clause, scope_node); + LABEL *label = NEW_LABEL(clause_lineno); + PUSH_LABEL(body_seq, label); - return; - } - case PM_CASE_NODE: { - // case foo; when bar; end - // ^^^^^^^^^^^^^^^^^^^^^^^ - const pm_case_node_t *cast = (const pm_case_node_t *) node; - const pm_node_list_t *conditions = &cast->conditions; + // Establish branch coverage for the when clause. + if (PM_BRANCH_COVERAGE_P(iseq)) { + rb_code_location_t branch_location = pm_code_location(scope_node, clause->statements != NULL ? ((const pm_node_t *) clause->statements) : ((const pm_node_t *) clause)); + add_trace_branch_coverage(iseq, body_seq, &branch_location, branch_location.beg_pos.column, branch_id++, "when", branches); + } - // This is the anchor that we will compile the conditions of the various - // `when` nodes into. If a match is found, they will need to jump into - // the body_seq anchor to the correct spot. - DECL_ANCHOR(cond_seq); - INIT_ANCHOR(cond_seq); + if (clause->statements != NULL) { + pm_compile_node(iseq, (const pm_node_t *) clause->statements, body_seq, popped, scope_node); + } + else if (!popped) { + PUSH_SYNTHETIC_PUTNIL(body_seq, iseq); + } - // This is the anchor that we will compile the bodies of the various - // `when` nodes into. We'll make sure that the clauses that are compiled - // jump into the correct spots within this anchor. - DECL_ANCHOR(body_seq); - INIT_ANCHOR(body_seq); + PUSH_INSNL(body_seq, location, jump, end_label); - // This is the label where all of the when clauses will jump to if they - // have matched and are done executing their bodies. - LABEL *end_label = NEW_LABEL(location.line); + // Compile each of the conditions for the when clause into the + // cond_seq. Each one should have a unique condition and should + // jump to the subsequent one if it doesn't match. + for (size_t condition_index = 0; condition_index < conditions->size; condition_index++) { + const pm_node_t *condition = conditions->nodes[condition_index]; - // If we have a predicate on this case statement, then it's going to - // compare all of the various when clauses to the predicate. If we - // don't, then it's basically an if-elsif-else chain. - if (cast->predicate == NULL) { - // Establish branch coverage for the case node. - VALUE branches = Qfalse; - rb_code_location_t case_location = { 0 }; - int branch_id = 0; + if (PM_NODE_TYPE_P(condition, PM_SPLAT_NODE)) { + pm_node_location_t cond_location = PM_NODE_START_LOCATION(condition); + PUSH_INSN(cond_seq, cond_location, putnil); + pm_compile_node(iseq, condition, cond_seq, false, scope_node); + PUSH_INSN1(cond_seq, cond_location, checkmatch, INT2FIX(VM_CHECKMATCH_TYPE_WHEN | VM_CHECKMATCH_ARRAY)); + PUSH_INSNL(cond_seq, cond_location, branchif, label); + } + else { + LABEL *next_label = NEW_LABEL(pm_node_line_number_cached(condition, scope_node)); + pm_compile_branch_condition(iseq, cond_seq, condition, label, next_label, scope_node); + PUSH_LABEL(cond_seq, next_label); + } + } + } - if (PM_BRANCH_COVERAGE_P(iseq)) { - case_location = pm_code_location(scope_node, (const pm_node_t *) cast); - branches = decl_branch_base(iseq, PTR2NUM(cast), &case_location, "case"); + // Establish branch coverage for the else clause (implicit or + // explicit). + if (PM_BRANCH_COVERAGE_P(iseq)) { + rb_code_location_t branch_location; + + if (cast->else_clause == NULL) { + branch_location = case_location; + } else if (cast->else_clause->statements == NULL) { + branch_location = pm_code_location(scope_node, (const pm_node_t *) cast->else_clause); + } else { + branch_location = pm_code_location(scope_node, (const pm_node_t *) cast->else_clause->statements); } - // Loop through each clauses in the case node and compile each of - // the conditions within them into cond_seq. If they match, they - // should jump into their respective bodies in body_seq. - for (size_t clause_index = 0; clause_index < conditions->size; clause_index++) { - const pm_when_node_t *clause = (const pm_when_node_t *) conditions->nodes[clause_index]; - const pm_node_list_t *conditions = &clause->conditions; + add_trace_branch_coverage(iseq, cond_seq, &branch_location, branch_location.beg_pos.column, branch_id, "else", branches); + } - int clause_lineno = pm_node_line_number(parser, (const pm_node_t *) clause); - LABEL *label = NEW_LABEL(clause_lineno); - PUSH_LABEL(body_seq, label); + // Compile the else clause if there is one. + if (cast->else_clause != NULL) { + pm_compile_node(iseq, (const pm_node_t *) cast->else_clause, cond_seq, popped, scope_node); + } + else if (!popped) { + PUSH_SYNTHETIC_PUTNIL(cond_seq, iseq); + } - // Establish branch coverage for the when clause. - if (PM_BRANCH_COVERAGE_P(iseq)) { - rb_code_location_t branch_location = pm_code_location(scope_node, clause->statements != NULL ? ((const pm_node_t *) clause->statements) : ((const pm_node_t *) clause)); - add_trace_branch_coverage(iseq, body_seq, &branch_location, branch_location.beg_pos.column, branch_id++, "when", branches); - } + // Finally, jump to the end label if none of the other conditions + // have matched. + PUSH_INSNL(cond_seq, location, jump, end_label); + PUSH_SEQ(ret, cond_seq); + } + else { + // Establish branch coverage for the case node. + VALUE branches = Qfalse; + rb_code_location_t case_location = { 0 }; + int branch_id = 0; - if (clause->statements != NULL) { - pm_compile_node(iseq, (const pm_node_t *) clause->statements, body_seq, popped, scope_node); - } - else if (!popped) { - PUSH_SYNTHETIC_PUTNIL(body_seq, iseq); - } + if (PM_BRANCH_COVERAGE_P(iseq)) { + case_location = pm_code_location(scope_node, (const pm_node_t *) cast); + branches = decl_branch_base(iseq, PTR2NUM(cast), &case_location, "case"); + } - PUSH_INSNL(body_seq, location, jump, end_label); + // This is the label where everything will fall into if none of the + // conditions matched. + LABEL *else_label = NEW_LABEL(location.line); - // Compile each of the conditions for the when clause into the - // cond_seq. Each one should have a unique condition and should - // jump to the subsequent one if it doesn't match. - for (size_t condition_index = 0; condition_index < conditions->size; condition_index++) { - const pm_node_t *condition = conditions->nodes[condition_index]; + // It's possible for us to speed up the case node by using a + // dispatch hash. This is a hash that maps the conditions of the + // various when clauses to the labels of their bodies. If we can + // compile the conditions into a hash key, then we can use a hash + // lookup to jump directly to the correct when clause body. + VALUE dispatch = Qundef; + if (ISEQ_COMPILE_DATA(iseq)->option->specialized_instruction) { + dispatch = cdhash_new(0); + } - if (PM_NODE_TYPE_P(condition, PM_SPLAT_NODE)) { - pm_node_location_t cond_location = PM_NODE_START_LOCATION(parser, condition); - PUSH_INSN(cond_seq, cond_location, putnil); - pm_compile_node(iseq, condition, cond_seq, false, scope_node); - PUSH_INSN1(cond_seq, cond_location, checkmatch, INT2FIX(VM_CHECKMATCH_TYPE_WHEN | VM_CHECKMATCH_ARRAY)); - PUSH_INSNL(cond_seq, cond_location, branchif, label); + // We're going to loop through each of the conditions in the case + // node and compile each of their contents into both the cond_seq + // and the body_seq. Each condition will use its own label to jump + // from its conditions into its body. + // + // Note that none of the code in the loop below should be adding + // anything to ret, as we're going to be laying out the entire case + // node instructions later. + for (size_t clause_index = 0; clause_index < conditions->size; clause_index++) { + const pm_when_node_t *clause = (const pm_when_node_t *) conditions->nodes[clause_index]; + pm_node_location_t clause_location = PM_NODE_START_LOCATION((const pm_node_t *) clause); + + const pm_node_list_t *conditions = &clause->conditions; + LABEL *label = NEW_LABEL(clause_location.line); + + // Compile each of the conditions for the when clause into the + // cond_seq. Each one should have a unique comparison that then + // jumps into the body if it matches. + for (size_t condition_index = 0; condition_index < conditions->size; condition_index++) { + const pm_node_t *condition = conditions->nodes[condition_index]; + const pm_node_location_t condition_location = PM_NODE_START_LOCATION(condition); + + // If we haven't already abandoned the optimization, then + // we're going to try to compile the condition into the + // dispatch hash. + if (dispatch != Qundef) { + dispatch = pm_compile_case_node_dispatch(iseq, dispatch, condition, label, scope_node); + } + + if (PM_NODE_TYPE_P(condition, PM_SPLAT_NODE)) { + PUSH_INSN(cond_seq, condition_location, dup); + pm_compile_node(iseq, condition, cond_seq, false, scope_node); + PUSH_INSN1(cond_seq, condition_location, checkmatch, INT2FIX(VM_CHECKMATCH_TYPE_CASE | VM_CHECKMATCH_ARRAY)); + } + else { + if (PM_NODE_TYPE_P(condition, PM_STRING_NODE)) { + const pm_string_node_t *string = (const pm_string_node_t *) condition; + VALUE value = parse_static_literal_string(iseq, scope_node, condition, &string->unescaped); + PUSH_INSN1(cond_seq, condition_location, putobject, value); } else { - LABEL *next_label = NEW_LABEL(pm_node_line_number(parser, condition)); - pm_compile_branch_condition(iseq, cond_seq, condition, label, next_label, false, scope_node); - PUSH_LABEL(cond_seq, next_label); + pm_compile_node(iseq, condition, cond_seq, false, scope_node); } + + PUSH_INSN1(cond_seq, condition_location, topn, INT2FIX(1)); + PUSH_SEND_WITH_FLAG(cond_seq, condition_location, idEqq, INT2NUM(1), INT2FIX(VM_CALL_FCALL | VM_CALL_ARGS_SIMPLE)); } - } - // Establish branch coverage for the else clause (implicit or - // explicit). - if (PM_BRANCH_COVERAGE_P(iseq)) { - rb_code_location_t branch_location; + PUSH_INSNL(cond_seq, condition_location, branchif, label); + } - if (cast->consequent == NULL) { - branch_location = case_location; - } else if (cast->consequent->statements == NULL) { - branch_location = pm_code_location(scope_node, (const pm_node_t *) cast->consequent); - } else { - branch_location = pm_code_location(scope_node, (const pm_node_t *) cast->consequent->statements); - } + // Now, add the label to the body and compile the body of the + // when clause. This involves popping the predicate, compiling + // the statements to be executed, and then compiling a jump to + // the end of the case node. + PUSH_LABEL(body_seq, label); + PUSH_INSN(body_seq, clause_location, pop); - add_trace_branch_coverage(iseq, cond_seq, &branch_location, branch_location.beg_pos.column, branch_id, "else", branches); + // Establish branch coverage for the when clause. + if (PM_BRANCH_COVERAGE_P(iseq)) { + rb_code_location_t branch_location = pm_code_location(scope_node, clause->statements != NULL ? ((const pm_node_t *) clause->statements) : ((const pm_node_t *) clause)); + add_trace_branch_coverage(iseq, body_seq, &branch_location, branch_location.beg_pos.column, branch_id++, "when", branches); } - // Compile the consequent else clause if there is one. - if (cast->consequent != NULL) { - pm_compile_node(iseq, (const pm_node_t *) cast->consequent, cond_seq, popped, scope_node); + if (clause->statements != NULL) { + pm_compile_node(iseq, (const pm_node_t *) clause->statements, body_seq, popped, scope_node); } else if (!popped) { - PUSH_SYNTHETIC_PUTNIL(cond_seq, iseq); + PUSH_SYNTHETIC_PUTNIL(body_seq, iseq); } - // Finally, jump to the end label if none of the other conditions - // have matched. - PUSH_INSNL(cond_seq, location, jump, end_label); - PUSH_SEQ(ret, cond_seq); + PUSH_INSNL(body_seq, clause_location, jump, end_label); } - else { - // Establish branch coverage for the case node. - VALUE branches = Qfalse; - rb_code_location_t case_location = { 0 }; - int branch_id = 0; + // Now that we have compiled the conditions and the bodies of the + // various when clauses, we can compile the predicate, lay out the + // conditions, compile the fallback subsequent if there is one, and + // finally put in the bodies of the when clauses. + PM_COMPILE_NOT_POPPED(cast->predicate); + + // If we have a dispatch hash, then we'll use it here to create the + // optimization. + if (dispatch != Qundef) { + PUSH_INSN(ret, location, dup); + RB_OBJ_SET_SHAREABLE(dispatch); // it is special that the hash is shareable but not frozen, because compile.c modify them. This Hahs instance is not accessible so it is safe to leave it. + PUSH_INSN2(ret, location, opt_case_dispatch, dispatch, else_label); + LABEL_REF(else_label); + } + + PUSH_SEQ(ret, cond_seq); + + // Compile either the explicit else clause or an implicit else + // clause. + PUSH_LABEL(ret, else_label); + + if (cast->else_clause != NULL) { + pm_node_location_t else_location = PM_NODE_START_LOCATION(cast->else_clause->statements != NULL ? ((const pm_node_t *) cast->else_clause->statements) : ((const pm_node_t *) cast->else_clause)); + PUSH_INSN(ret, else_location, pop); + + // Establish branch coverage for the else clause. if (PM_BRANCH_COVERAGE_P(iseq)) { - case_location = pm_code_location(scope_node, (const pm_node_t *) cast); - branches = decl_branch_base(iseq, PTR2NUM(cast), &case_location, "case"); + rb_code_location_t branch_location = pm_code_location(scope_node, cast->else_clause->statements != NULL ? ((const pm_node_t *) cast->else_clause->statements) : ((const pm_node_t *) cast->else_clause)); + add_trace_branch_coverage(iseq, ret, &branch_location, branch_location.beg_pos.column, branch_id, "else", branches); } - // This is the label where everything will fall into if none of the - // conditions matched. - LABEL *else_label = NEW_LABEL(location.line); + PM_COMPILE((const pm_node_t *) cast->else_clause); + PUSH_INSNL(ret, else_location, jump, end_label); + } + else { + PUSH_INSN(ret, location, pop); - // It's possible for us to speed up the case node by using a - // dispatch hash. This is a hash that maps the conditions of the - // various when clauses to the labels of their bodies. If we can - // compile the conditions into a hash key, then we can use a hash - // lookup to jump directly to the correct when clause body. - VALUE dispatch = Qundef; - if (ISEQ_COMPILE_DATA(iseq)->option->specialized_instruction) { - dispatch = rb_hash_new(); - RHASH_TBL_RAW(dispatch)->type = &cdhash_type; + // Establish branch coverage for the implicit else clause. + if (PM_BRANCH_COVERAGE_P(iseq)) { + add_trace_branch_coverage(iseq, ret, &case_location, case_location.beg_pos.column, branch_id, "else", branches); } - // We're going to loop through each of the conditions in the case - // node and compile each of their contents into both the cond_seq - // and the body_seq. Each condition will use its own label to jump - // from its conditions into its body. - // - // Note that none of the code in the loop below should be adding - // anything to ret, as we're going to be laying out the entire case - // node instructions later. - for (size_t clause_index = 0; clause_index < conditions->size; clause_index++) { - const pm_when_node_t *clause = (const pm_when_node_t *) conditions->nodes[clause_index]; - pm_node_location_t clause_location = PM_NODE_START_LOCATION(parser, (const pm_node_t *) clause); - - const pm_node_list_t *conditions = &clause->conditions; - LABEL *label = NEW_LABEL(clause_location.line); - - // Compile each of the conditions for the when clause into the - // cond_seq. Each one should have a unique comparison that then - // jumps into the body if it matches. - for (size_t condition_index = 0; condition_index < conditions->size; condition_index++) { - const pm_node_t *condition = conditions->nodes[condition_index]; - const pm_node_location_t condition_location = PM_NODE_START_LOCATION(parser, condition); - - // If we haven't already abandoned the optimization, then - // we're going to try to compile the condition into the - // dispatch hash. - if (dispatch != Qundef) { - dispatch = pm_compile_case_node_dispatch(iseq, dispatch, condition, label, scope_node); - } + if (!popped) PUSH_INSN(ret, location, putnil); + PUSH_INSNL(ret, location, jump, end_label); + } + } - if (PM_NODE_TYPE_P(condition, PM_SPLAT_NODE)) { - PUSH_INSN(cond_seq, condition_location, dup); - pm_compile_node(iseq, condition, cond_seq, false, scope_node); - PUSH_INSN1(cond_seq, condition_location, checkmatch, INT2FIX(VM_CHECKMATCH_TYPE_CASE | VM_CHECKMATCH_ARRAY)); - } - else { - if (PM_NODE_TYPE_P(condition, PM_STRING_NODE)) { - const pm_string_node_t *string = (const pm_string_node_t *) condition; - VALUE value = parse_static_literal_string(iseq, scope_node, condition, &string->unescaped); - PUSH_INSN1(cond_seq, condition_location, putobject, value); - } - else { - pm_compile_node(iseq, condition, cond_seq, false, scope_node); - } + PUSH_SEQ(ret, body_seq); + PUSH_LABEL(ret, end_label); +} - PUSH_INSN1(cond_seq, condition_location, topn, INT2FIX(1)); - PUSH_SEND_WITH_FLAG(cond_seq, condition_location, idEqq, INT2NUM(1), INT2FIX(VM_CALL_FCALL | VM_CALL_ARGS_SIMPLE)); - } +static inline void +pm_compile_case_match_node(rb_iseq_t *iseq, const pm_case_match_node_t *node, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + // This is the anchor that we will compile the bodies of the various + // `in` nodes into. We'll make sure that the patterns that are compiled + // jump into the correct spots within this anchor. + DECL_ANCHOR(body_seq); + + // This is the anchor that we will compile the patterns of the various + // `in` nodes into. If a match is found, they will need to jump into the + // body_seq anchor to the correct spot. + DECL_ANCHOR(cond_seq); - PUSH_INSNL(cond_seq, condition_location, branchif, label); - } + // This label is used to indicate the end of the entire node. It is + // jumped to after the entire stack is cleaned up. + LABEL *end_label = NEW_LABEL(location->line); - // Now, add the label to the body and compile the body of the - // when clause. This involves popping the predicate, compiling - // the statements to be executed, and then compiling a jump to - // the end of the case node. - PUSH_LABEL(body_seq, label); - PUSH_INSN(body_seq, clause_location, pop); - - // Establish branch coverage for the when clause. - if (PM_BRANCH_COVERAGE_P(iseq)) { - rb_code_location_t branch_location = pm_code_location(scope_node, clause->statements != NULL ? ((const pm_node_t *) clause->statements) : ((const pm_node_t *) clause)); - add_trace_branch_coverage(iseq, body_seq, &branch_location, branch_location.beg_pos.column, branch_id++, "when", branches); - } + // This label is used as the fallback for the case match. If no match is + // found, then we jump to this label. This is either an `else` clause or + // an error handler. + LABEL *else_label = NEW_LABEL(location->line); - if (clause->statements != NULL) { - pm_compile_node(iseq, (const pm_node_t *) clause->statements, body_seq, popped, scope_node); - } - else if (!popped) { - PUSH_SYNTHETIC_PUTNIL(body_seq, iseq); - } + // We're going to use this to uniquely identify each branch so that we + // can track coverage information. + rb_code_location_t case_location = { 0 }; + VALUE branches = Qfalse; + int branch_id = 0; - PUSH_INSNL(body_seq, clause_location, jump, end_label); - } + if (PM_BRANCH_COVERAGE_P(iseq)) { + case_location = pm_code_location(scope_node, (const pm_node_t *) node); + branches = decl_branch_base(iseq, PTR2NUM(node), &case_location, "case"); + } - // Now that we have compiled the conditions and the bodies of the - // various when clauses, we can compile the predicate, lay out the - // conditions, compile the fallback consequent if there is one, and - // finally put in the bodies of the when clauses. - PM_COMPILE_NOT_POPPED(cast->predicate); + // If there is only one pattern, then the behavior changes a bit. It + // effectively gets treated as a match required node (this is how it is + // represented in the other parser). + bool in_single_pattern = node->else_clause == NULL && node->conditions.size == 1; - // If we have a dispatch hash, then we'll use it here to create the - // optimization. - if (dispatch != Qundef) { - PUSH_INSN(ret, location, dup); - PUSH_INSN2(ret, location, opt_case_dispatch, dispatch, else_label); - LABEL_REF(else_label); - } + // First, we're going to push a bunch of stuff onto the stack that is + // going to serve as our scratch space. + if (in_single_pattern) { + PUSH_INSN(ret, *location, putnil); // key error key + PUSH_INSN(ret, *location, putnil); // key error matchee + PUSH_INSN1(ret, *location, putobject, Qfalse); // key error? + PUSH_INSN(ret, *location, putnil); // error string + } - PUSH_SEQ(ret, cond_seq); + // Now we're going to compile the value to match against. + PUSH_INSN(ret, *location, putnil); // deconstruct cache + PM_COMPILE_NOT_POPPED(node->predicate); - // Compile either the explicit else clause or an implicit else - // clause. - PUSH_LABEL(ret, else_label); + // Next, we'll loop through every in clause and compile its body into + // the body_seq anchor and its pattern into the cond_seq anchor. We'll + // make sure the pattern knows how to jump correctly into the body if it + // finds a match. + for (size_t index = 0; index < node->conditions.size; index++) { + const pm_node_t *condition = node->conditions.nodes[index]; + RUBY_ASSERT(PM_NODE_TYPE_P(condition, PM_IN_NODE)); - if (cast->consequent != NULL) { - pm_node_location_t else_location = PM_NODE_START_LOCATION(parser, cast->consequent->statements != NULL ? ((const pm_node_t *) cast->consequent->statements) : ((const pm_node_t *) cast->consequent)); - PUSH_INSN(ret, else_location, pop); + const pm_in_node_t *in_node = (const pm_in_node_t *) condition; + const pm_node_location_t in_location = PM_NODE_START_LOCATION(in_node); + const pm_node_location_t pattern_location = PM_NODE_START_LOCATION(in_node->pattern); - // Establish branch coverage for the else clause. - if (PM_BRANCH_COVERAGE_P(iseq)) { - rb_code_location_t branch_location = pm_code_location(scope_node, cast->consequent->statements != NULL ? ((const pm_node_t *) cast->consequent->statements) : ((const pm_node_t *) cast->consequent)); - add_trace_branch_coverage(iseq, ret, &branch_location, branch_location.beg_pos.column, branch_id, "else", branches); - } + if (branch_id) { + PUSH_INSN(body_seq, in_location, putnil); + } - PM_COMPILE((const pm_node_t *) cast->consequent); - PUSH_INSNL(ret, else_location, jump, end_label); - } - else { - PUSH_INSN(ret, location, pop); + LABEL *body_label = NEW_LABEL(in_location.line); + PUSH_LABEL(body_seq, body_label); + PUSH_INSN1(body_seq, in_location, adjuststack, INT2FIX(in_single_pattern ? 6 : 2)); - // Establish branch coverage for the implicit else clause. - if (PM_BRANCH_COVERAGE_P(iseq)) { - add_trace_branch_coverage(iseq, ret, &case_location, case_location.beg_pos.column, branch_id, "else", branches); - } + // Establish branch coverage for the in clause. + if (PM_BRANCH_COVERAGE_P(iseq)) { + rb_code_location_t branch_location = pm_code_location(scope_node, in_node->statements != NULL ? ((const pm_node_t *) in_node->statements) : ((const pm_node_t *) in_node)); + add_trace_branch_coverage(iseq, body_seq, &branch_location, branch_location.beg_pos.column, branch_id++, "in", branches); + } + + if (in_node->statements != NULL) { + PM_COMPILE_INTO_ANCHOR(body_seq, (const pm_node_t *) in_node->statements); + } + else if (!popped) { + PUSH_SYNTHETIC_PUTNIL(body_seq, iseq); + } + + PUSH_INSNL(body_seq, in_location, jump, end_label); + LABEL *next_pattern_label = NEW_LABEL(pattern_location.line); + + PUSH_INSN(cond_seq, pattern_location, dup); + pm_compile_pattern(iseq, scope_node, in_node->pattern, cond_seq, body_label, next_pattern_label, in_single_pattern, true, 2); + PUSH_LABEL(cond_seq, next_pattern_label); + LABEL_UNREMOVABLE(next_pattern_label); + } + + if (node->else_clause != NULL) { + // If we have an `else` clause, then this becomes our fallback (and + // there is no need to compile in code to potentially raise an + // error). + const pm_else_node_t *else_node = node->else_clause; + + PUSH_LABEL(cond_seq, else_label); + PUSH_INSN(cond_seq, *location, pop); + PUSH_INSN(cond_seq, *location, pop); + + // Establish branch coverage for the else clause. + if (PM_BRANCH_COVERAGE_P(iseq)) { + rb_code_location_t branch_location = pm_code_location(scope_node, else_node->statements != NULL ? ((const pm_node_t *) else_node->statements) : ((const pm_node_t *) else_node)); + add_trace_branch_coverage(iseq, cond_seq, &branch_location, branch_location.beg_pos.column, branch_id, "else", branches); + } + + PM_COMPILE_INTO_ANCHOR(cond_seq, (const pm_node_t *) else_node); + PUSH_INSNL(cond_seq, *location, jump, end_label); + PUSH_INSN(cond_seq, *location, putnil); + if (popped) PUSH_INSN(cond_seq, *location, putnil); + } + else { + // Otherwise, if we do not have an `else` clause, we will compile in + // the code to handle raising an appropriate error. + PUSH_LABEL(cond_seq, else_label); + + // Establish branch coverage for the implicit else clause. + add_trace_branch_coverage(iseq, cond_seq, &case_location, case_location.beg_pos.column, branch_id, "else", branches); + + if (in_single_pattern) { + pm_compile_pattern_error_handler(iseq, scope_node, (const pm_node_t *) node, cond_seq, end_label, popped); + } + else { + PUSH_INSN1(cond_seq, *location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); + PUSH_INSN1(cond_seq, *location, putobject, rb_eNoMatchingPatternError); + PUSH_INSN1(cond_seq, *location, topn, INT2FIX(2)); + PUSH_SEND(cond_seq, *location, id_core_raise, INT2FIX(2)); + + PUSH_INSN1(cond_seq, *location, adjuststack, INT2FIX(3)); + if (!popped) PUSH_INSN(cond_seq, *location, putnil); + PUSH_INSNL(cond_seq, *location, jump, end_label); + PUSH_INSN1(cond_seq, *location, dupn, INT2FIX(1)); + if (popped) PUSH_INSN(cond_seq, *location, putnil); + } + } + + // At the end of all of this compilation, we will add the code for the + // conditions first, then the various bodies, then mark the end of the + // entire sequence with the end label. + PUSH_SEQ(ret, cond_seq); + PUSH_SEQ(ret, body_seq); + PUSH_LABEL(ret, end_label); +} + +static inline void +pm_compile_forwarding_super_node(rb_iseq_t *iseq, const pm_forwarding_super_node_t *node, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + const rb_iseq_t *block = NULL; + const rb_iseq_t *previous_block = NULL; + LABEL *retry_label = NULL; + LABEL *retry_end_l = NULL; + + if (node->block != NULL) { + previous_block = ISEQ_COMPILE_DATA(iseq)->current_block; + ISEQ_COMPILE_DATA(iseq)->current_block = NULL; + + retry_label = NEW_LABEL(location->line); + retry_end_l = NEW_LABEL(location->line); + + PUSH_LABEL(ret, retry_label); + } + else { + iseq_set_use_block(ISEQ_BODY(iseq)->local_iseq); + } + + PUSH_INSN(ret, *location, putself); + int flag = VM_CALL_ZSUPER | VM_CALL_SUPER | VM_CALL_FCALL; + + if (node->block != NULL) { + pm_scope_node_t next_scope_node; + pm_scope_node_init((const pm_node_t *) node->block, &next_scope_node, scope_node); + + ISEQ_COMPILE_DATA(iseq)->current_block = block = NEW_CHILD_ISEQ(&next_scope_node, make_name_for_block(iseq), ISEQ_TYPE_BLOCK, location->line); + pm_scope_node_destroy(&next_scope_node); + RB_OBJ_WRITTEN(iseq, Qundef, (VALUE) block); + } + + DECL_ANCHOR(args); + + struct rb_iseq_constant_body *const body = ISEQ_BODY(iseq); + const rb_iseq_t *local_iseq = body->local_iseq; + const struct rb_iseq_constant_body *const local_body = ISEQ_BODY(local_iseq); + + int argc = 0; + int depth = get_lvar_level(iseq); + + if (ISEQ_BODY(ISEQ_BODY(iseq)->local_iseq)->param.flags.forwardable) { + flag |= VM_CALL_FORWARDING; + pm_local_index_t mult_local = pm_lookup_local_index(iseq, scope_node, PM_CONSTANT_DOT3, 0); + PUSH_GETLOCAL(ret, *location, mult_local.index, mult_local.level); + + const struct rb_callinfo *callinfo = new_callinfo(iseq, 0, 0, flag, NULL, block != NULL); + PUSH_INSN2(ret, *location, invokesuperforward, callinfo, block); + + if (popped) PUSH_INSN(ret, *location, pop); + if (node->block) { + ISEQ_COMPILE_DATA(iseq)->current_block = previous_block; + } + return; + } + + if (local_body->param.flags.has_lead) { + /* required arguments */ + for (int i = 0; i < local_body->param.lead_num; i++) { + int idx = local_body->local_table_size - i; + PUSH_GETLOCAL(args, *location, idx, depth); + } + argc += local_body->param.lead_num; + } + + if (local_body->param.flags.has_opt) { + /* optional arguments */ + for (int j = 0; j < local_body->param.opt_num; j++) { + int idx = local_body->local_table_size - (argc + j); + PUSH_GETLOCAL(args, *location, idx, depth); + } + argc += local_body->param.opt_num; + } + + if (local_body->param.flags.has_rest) { + /* rest argument */ + int idx = local_body->local_table_size - local_body->param.rest_start; + PUSH_GETLOCAL(args, *location, idx, depth); + PUSH_INSN1(args, *location, splatarray, Qfalse); + + argc = local_body->param.rest_start + 1; + flag |= VM_CALL_ARGS_SPLAT; + } + + if (local_body->param.flags.has_post) { + /* post arguments */ + int post_len = local_body->param.post_num; + int post_start = local_body->param.post_start; + + int j = 0; + for (; j < post_len; j++) { + int idx = local_body->local_table_size - (post_start + j); + PUSH_GETLOCAL(args, *location, idx, depth); + } + + if (local_body->param.flags.has_rest) { + // argc remains unchanged from rest branch + PUSH_INSN1(args, *location, newarray, INT2FIX(j)); + PUSH_INSN(args, *location, concatarray); + } + else { + argc = post_len + post_start; + } + } + + const struct rb_iseq_param_keyword *const local_keyword = local_body->param.keyword; + if (local_body->param.flags.has_kw) { + int local_size = local_body->local_table_size; + argc++; + + PUSH_INSN1(args, *location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); - if (!popped) PUSH_INSN(ret, location, putnil); - PUSH_INSNL(ret, location, jump, end_label); + if (local_body->param.flags.has_kwrest) { + int idx = local_body->local_table_size - local_keyword->rest_start; + PUSH_GETLOCAL(args, *location, idx, depth); + RUBY_ASSERT(local_keyword->num > 0); + PUSH_SEND(args, *location, rb_intern("dup"), INT2FIX(0)); + } + else { + PUSH_INSN1(args, *location, newhash, INT2FIX(0)); + } + int i = 0; + for (; i < local_keyword->num; ++i) { + ID id = local_keyword->table[i]; + int idx = local_size - get_local_var_idx(local_iseq, id); + + { + VALUE operand = ID2SYM(id); + PUSH_INSN1(args, *location, putobject, operand); } + + PUSH_GETLOCAL(args, *location, idx, depth); } - PUSH_SEQ(ret, body_seq); - PUSH_LABEL(ret, end_label); + PUSH_SEND(args, *location, id_core_hash_merge_ptr, INT2FIX(i * 2 + 1)); + flag |= VM_CALL_KW_SPLAT| VM_CALL_KW_SPLAT_MUT; + } + else if (local_body->param.flags.has_kwrest) { + int idx = local_body->local_table_size - local_keyword->rest_start; + PUSH_GETLOCAL(args, *location, idx, depth); + argc++; + flag |= VM_CALL_KW_SPLAT; + } + + PUSH_SEQ(ret, args); + + { + const struct rb_callinfo *callinfo = new_callinfo(iseq, 0, argc, flag, NULL, block != NULL); + PUSH_INSN2(ret, *location, invokesuper, callinfo, block); + } + + if (node->block != NULL) { + pm_compile_retry_end_label(iseq, ret, retry_end_l); + PUSH_CATCH_ENTRY(CATCH_TYPE_BREAK, retry_label, retry_end_l, block, retry_end_l); + ISEQ_COMPILE_DATA(iseq)->current_block = previous_block; + } + + if (popped) PUSH_INSN(ret, *location, pop); +} + +static inline void +pm_compile_match_required_node(rb_iseq_t *iseq, const pm_match_required_node_t *node, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + LABEL *matched_label = NEW_LABEL(location->line); + LABEL *unmatched_label = NEW_LABEL(location->line); + LABEL *done_label = NEW_LABEL(location->line); + + // First, we're going to push a bunch of stuff onto the stack that is + // going to serve as our scratch space. + PUSH_INSN(ret, *location, putnil); // key error key + PUSH_INSN(ret, *location, putnil); // key error matchee + PUSH_INSN1(ret, *location, putobject, Qfalse); // key error? + PUSH_INSN(ret, *location, putnil); // error string + PUSH_INSN(ret, *location, putnil); // deconstruct cache + + // Next we're going to compile the value expression such that it's on + // the stack. + PM_COMPILE_NOT_POPPED(node->value); + + // Here we'll dup it so that it can be used for comparison, but also be + // used for error handling. + PUSH_INSN(ret, *location, dup); + + // Next we'll compile the pattern. We indicate to the pm_compile_pattern + // function that this is the only pattern that will be matched against + // through the in_single_pattern parameter. We also indicate that the + // value to compare against is 2 slots from the top of the stack (the + // base_index parameter). + pm_compile_pattern(iseq, scope_node, node->pattern, ret, matched_label, unmatched_label, true, true, 2); + + // If the pattern did not match the value, then we're going to compile + // in our error handler code. This will determine which error to raise + // and raise it. + PUSH_LABEL(ret, unmatched_label); + pm_compile_pattern_error_handler(iseq, scope_node, (const pm_node_t *) node, ret, done_label, popped); + + // If the pattern did match, we'll clean up the values we've pushed onto + // the stack and then push nil onto the stack if it's not popped. + PUSH_LABEL(ret, matched_label); + PUSH_INSN1(ret, *location, adjuststack, INT2FIX(6)); + if (!popped) PUSH_INSN(ret, *location, putnil); + PUSH_INSNL(ret, *location, jump, done_label); + PUSH_LABEL(ret, done_label); +} + +static inline void +pm_compile_match_write_node(rb_iseq_t *iseq, const pm_match_write_node_t *node, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + LABEL *fail_label = NEW_LABEL(location->line); + LABEL *end_label = NEW_LABEL(location->line); + + // First, we'll compile the call so that all of its instructions are + // present. Then we'll compile all of the local variable targets. + PM_COMPILE_NOT_POPPED((const pm_node_t *) node->call); + + // Now, check if the match was successful. If it was, then we'll + // continue on and assign local variables. Otherwise we'll skip over the + // assignment code. + { + VALUE operand = rb_id2sym(idBACKREF); + PUSH_INSN1(ret, *location, getglobal, operand); + } + + PUSH_INSN(ret, *location, dup); + PUSH_INSNL(ret, *location, branchunless, fail_label); + + // If there's only a single local variable target, we can skip some of + // the bookkeeping, so we'll put a special branch here. + size_t targets_count = node->targets.size; + + if (targets_count == 1) { + const pm_node_t *target = node->targets.nodes[0]; + RUBY_ASSERT(PM_NODE_TYPE_P(target, PM_LOCAL_VARIABLE_TARGET_NODE)); + + const pm_local_variable_target_node_t *local_target = (const pm_local_variable_target_node_t *) target; + pm_local_index_t index = pm_lookup_local_index(iseq, scope_node, local_target->name, local_target->depth); + + { + VALUE operand = rb_id2sym(pm_constant_id_lookup(scope_node, local_target->name)); + PUSH_INSN1(ret, *location, putobject, operand); + } + + PUSH_SEND(ret, *location, idAREF, INT2FIX(1)); + PUSH_LABEL(ret, fail_label); + PUSH_SETLOCAL(ret, *location, index.index, index.level); + if (popped) PUSH_INSN(ret, *location, pop); return; - } - case PM_CASE_MATCH_NODE: { - // case foo; in bar; end - // ^^^^^^^^^^^^^^^^^^^^^ - // - // If you use the `case` keyword to create a case match node, it will - // match against all of the `in` clauses until it finds one that - // matches. If it doesn't find one, it can optionally fall back to an - // `else` clause. If none is present and a match wasn't found, it will - // raise an appropriate error. - const pm_case_match_node_t *cast = (const pm_case_match_node_t *) node; + } - // This is the anchor that we will compile the bodies of the various - // `in` nodes into. We'll make sure that the patterns that are compiled - // jump into the correct spots within this anchor. - DECL_ANCHOR(body_seq); - INIT_ANCHOR(body_seq); + DECL_ANCHOR(fail_anchor); - // This is the anchor that we will compile the patterns of the various - // `in` nodes into. If a match is found, they will need to jump into the - // body_seq anchor to the correct spot. - DECL_ANCHOR(cond_seq); - INIT_ANCHOR(cond_seq); + // Otherwise there is more than one local variable target, so we'll need + // to do some bookkeeping. + for (size_t targets_index = 0; targets_index < targets_count; targets_index++) { + const pm_node_t *target = node->targets.nodes[targets_index]; + RUBY_ASSERT(PM_NODE_TYPE_P(target, PM_LOCAL_VARIABLE_TARGET_NODE)); - // This label is used to indicate the end of the entire node. It is - // jumped to after the entire stack is cleaned up. - LABEL *end_label = NEW_LABEL(location.line); + const pm_local_variable_target_node_t *local_target = (const pm_local_variable_target_node_t *) target; + pm_local_index_t index = pm_lookup_local_index(iseq, scope_node, local_target->name, local_target->depth); - // This label is used as the fallback for the case match. If no match is - // found, then we jump to this label. This is either an `else` clause or - // an error handler. - LABEL *else_label = NEW_LABEL(location.line); + if (((size_t) targets_index) < (targets_count - 1)) { + PUSH_INSN(ret, *location, dup); + } - // We're going to use this to uniquely identify each branch so that we - // can track coverage information. - rb_code_location_t case_location = { 0 }; - VALUE branches = Qfalse; - int branch_id = 0; + { + VALUE operand = rb_id2sym(pm_constant_id_lookup(scope_node, local_target->name)); + PUSH_INSN1(ret, *location, putobject, operand); + } - if (PM_BRANCH_COVERAGE_P(iseq)) { - case_location = pm_code_location(scope_node, (const pm_node_t *) cast); - branches = decl_branch_base(iseq, PTR2NUM(cast), &case_location, "case"); + PUSH_SEND(ret, *location, idAREF, INT2FIX(1)); + PUSH_SETLOCAL(ret, *location, index.index, index.level); + + PUSH_INSN(fail_anchor, *location, putnil); + PUSH_SETLOCAL(fail_anchor, *location, index.index, index.level); + } + + // Since we matched successfully, now we'll jump to the end. + PUSH_INSNL(ret, *location, jump, end_label); + + // In the case that the match failed, we'll loop through each local + // variable target and set all of them to `nil`. + PUSH_LABEL(ret, fail_label); + PUSH_INSN(ret, *location, pop); + PUSH_SEQ(ret, fail_anchor); + + // Finally, we can push the end label for either case. + PUSH_LABEL(ret, end_label); + if (popped) PUSH_INSN(ret, *location, pop); +} + +static inline void +pm_compile_next_node(rb_iseq_t *iseq, const pm_next_node_t *node, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + if (ISEQ_COMPILE_DATA(iseq)->redo_label != 0 && can_add_ensure_iseq(iseq)) { + LABEL *splabel = NEW_LABEL(0); + PUSH_LABEL(ret, splabel); + + if (node->arguments) { + PM_COMPILE_NOT_POPPED((const pm_node_t *) node->arguments); + } + else { + PUSH_INSN(ret, *location, putnil); } + pm_add_ensure_iseq(ret, iseq, 0, scope_node); - // If there is only one pattern, then the behavior changes a bit. It - // effectively gets treated as a match required node (this is how it is - // represented in the other parser). - bool in_single_pattern = cast->consequent == NULL && cast->conditions.size == 1; + PUSH_ADJUST(ret, *location, ISEQ_COMPILE_DATA(iseq)->redo_label); + PUSH_INSNL(ret, *location, jump, ISEQ_COMPILE_DATA(iseq)->start_label); - // First, we're going to push a bunch of stuff onto the stack that is - // going to serve as our scratch space. - if (in_single_pattern) { - PUSH_INSN(ret, location, putnil); // key error key - PUSH_INSN(ret, location, putnil); // key error matchee - PUSH_INSN1(ret, location, putobject, Qfalse); // key error? - PUSH_INSN(ret, location, putnil); // error string + PUSH_ADJUST_RESTORE(ret, splabel); + if (!popped) PUSH_INSN(ret, *location, putnil); + } + else if (ISEQ_COMPILE_DATA(iseq)->end_label && can_add_ensure_iseq(iseq)) { + LABEL *splabel = NEW_LABEL(0); + + PUSH_LABEL(ret, splabel); + PUSH_ADJUST(ret, *location, ISEQ_COMPILE_DATA(iseq)->start_label); + + if (node->arguments != NULL) { + PM_COMPILE_NOT_POPPED((const pm_node_t *) node->arguments); + } + else { + PUSH_INSN(ret, *location, putnil); } - // Now we're going to compile the value to match against. - PUSH_INSN(ret, location, putnil); // deconstruct cache - PM_COMPILE_NOT_POPPED(cast->predicate); + pm_add_ensure_iseq(ret, iseq, 0, scope_node); + PUSH_INSNL(ret, *location, jump, ISEQ_COMPILE_DATA(iseq)->end_label); + PUSH_ADJUST_RESTORE(ret, splabel); + splabel->unremovable = FALSE; - // Next, we'll loop through every in clause and compile its body into - // the body_seq anchor and its pattern into the cond_seq anchor. We'll - // make sure the pattern knows how to jump correctly into the body if it - // finds a match. - for (size_t index = 0; index < cast->conditions.size; index++) { - const pm_node_t *condition = cast->conditions.nodes[index]; - RUBY_ASSERT(PM_NODE_TYPE_P(condition, PM_IN_NODE)); + if (!popped) PUSH_INSN(ret, *location, putnil); + } + else { + const rb_iseq_t *ip = iseq; + unsigned long throw_flag = 0; - const pm_in_node_t *in_node = (const pm_in_node_t *) condition; - const pm_node_location_t in_location = PM_NODE_START_LOCATION(parser, in_node); - const pm_node_location_t pattern_location = PM_NODE_START_LOCATION(parser, in_node->pattern); + while (ip) { + if (!ISEQ_COMPILE_DATA(ip)) { + ip = 0; + break; + } - if (branch_id) { - PUSH_INSN(body_seq, in_location, putnil); + throw_flag = VM_THROW_NO_ESCAPE_FLAG; + if (ISEQ_COMPILE_DATA(ip)->redo_label != 0) { + /* while loop */ + break; + } + else if (ISEQ_BODY(ip)->type == ISEQ_TYPE_BLOCK) { + break; + } + else if (ISEQ_BODY(ip)->type == ISEQ_TYPE_EVAL) { + COMPILE_ERROR(iseq, location->line, "Invalid next"); + return; } - LABEL *body_label = NEW_LABEL(in_location.line); - PUSH_LABEL(body_seq, body_label); - PUSH_INSN1(body_seq, in_location, adjuststack, INT2FIX(in_single_pattern ? 6 : 2)); + ip = ISEQ_BODY(ip)->parent_iseq; + } - // Establish branch coverage for the in clause. - if (PM_BRANCH_COVERAGE_P(iseq)) { - rb_code_location_t branch_location = pm_code_location(scope_node, in_node->statements != NULL ? ((const pm_node_t *) in_node->statements) : ((const pm_node_t *) in_node)); - add_trace_branch_coverage(iseq, body_seq, &branch_location, branch_location.beg_pos.column, branch_id++, "in", branches); + if (ip != 0) { + if (node->arguments) { + PM_COMPILE_NOT_POPPED((const pm_node_t *) node->arguments); + } + else { + PUSH_INSN(ret, *location, putnil); + } + + PUSH_INSN1(ret, *location, throw, INT2FIX(throw_flag | TAG_NEXT)); + if (popped) PUSH_INSN(ret, *location, pop); + } + else { + COMPILE_ERROR(iseq, location->line, "Invalid next"); + } + } +} + +static inline void +pm_compile_redo_node(rb_iseq_t *iseq, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + if (ISEQ_COMPILE_DATA(iseq)->redo_label && can_add_ensure_iseq(iseq)) { + LABEL *splabel = NEW_LABEL(0); + + PUSH_LABEL(ret, splabel); + PUSH_ADJUST(ret, *location, ISEQ_COMPILE_DATA(iseq)->redo_label); + pm_add_ensure_iseq(ret, iseq, 0, scope_node); + + PUSH_INSNL(ret, *location, jump, ISEQ_COMPILE_DATA(iseq)->redo_label); + PUSH_ADJUST_RESTORE(ret, splabel); + if (!popped) PUSH_INSN(ret, *location, putnil); + } + else if (ISEQ_BODY(iseq)->type != ISEQ_TYPE_EVAL && ISEQ_COMPILE_DATA(iseq)->start_label && can_add_ensure_iseq(iseq)) { + LABEL *splabel = NEW_LABEL(0); + + PUSH_LABEL(ret, splabel); + pm_add_ensure_iseq(ret, iseq, 0, scope_node); + PUSH_ADJUST(ret, *location, ISEQ_COMPILE_DATA(iseq)->start_label); + + PUSH_INSNL(ret, *location, jump, ISEQ_COMPILE_DATA(iseq)->start_label); + PUSH_ADJUST_RESTORE(ret, splabel); + if (!popped) PUSH_INSN(ret, *location, putnil); + } + else { + const rb_iseq_t *ip = iseq; + + while (ip) { + if (!ISEQ_COMPILE_DATA(ip)) { + ip = 0; + break; } - if (in_node->statements != NULL) { - PM_COMPILE_INTO_ANCHOR(body_seq, (const pm_node_t *) in_node->statements); + if (ISEQ_COMPILE_DATA(ip)->redo_label != 0) { + break; } - else if (!popped) { - PUSH_SYNTHETIC_PUTNIL(body_seq, iseq); + else if (ISEQ_BODY(ip)->type == ISEQ_TYPE_BLOCK) { + break; + } + else if (ISEQ_BODY(ip)->type == ISEQ_TYPE_EVAL) { + COMPILE_ERROR(iseq, location->line, "Invalid redo"); + return; } - PUSH_INSNL(body_seq, in_location, jump, end_label); - LABEL *next_pattern_label = NEW_LABEL(pattern_location.line); + ip = ISEQ_BODY(ip)->parent_iseq; + } - PUSH_INSN(cond_seq, pattern_location, dup); - pm_compile_pattern(iseq, scope_node, in_node->pattern, cond_seq, body_label, next_pattern_label, in_single_pattern, false, true, 2); - PUSH_LABEL(cond_seq, next_pattern_label); - LABEL_UNREMOVABLE(next_pattern_label); + if (ip != 0) { + PUSH_INSN(ret, *location, putnil); + PUSH_INSN1(ret, *location, throw, INT2FIX(VM_THROW_NO_ESCAPE_FLAG | TAG_REDO)); + if (popped) PUSH_INSN(ret, *location, pop); } + else { + COMPILE_ERROR(iseq, location->line, "Invalid redo"); + } + } +} + +static inline void +pm_compile_rescue_node(rb_iseq_t *iseq, const pm_rescue_node_t *node, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + iseq_set_exception_local_table(iseq); - if (cast->consequent != NULL) { - // If we have an `else` clause, then this becomes our fallback (and - // there is no need to compile in code to potentially raise an - // error). - const pm_else_node_t *else_node = (const pm_else_node_t *) cast->consequent; + // First, establish the labels that we need to be able to jump to within + // this compilation block. + LABEL *exception_match_label = NEW_LABEL(location->line); + LABEL *rescue_end_label = NEW_LABEL(location->line); - PUSH_LABEL(cond_seq, else_label); - PUSH_INSN(cond_seq, location, pop); - PUSH_INSN(cond_seq, location, pop); + // Next, compile each of the exceptions that we're going to be + // handling. For each one, we'll add instructions to check if the + // exception matches the raised one, and if it does then jump to the + // exception_match_label label. Otherwise it will fall through to the + // subsequent check. If there are no exceptions, we'll only check + // StandardError. + const pm_node_list_t *exceptions = &node->exceptions; - // Establish branch coverage for the else clause. - if (PM_BRANCH_COVERAGE_P(iseq)) { - rb_code_location_t branch_location = pm_code_location(scope_node, else_node->statements != NULL ? ((const pm_node_t *) else_node->statements) : ((const pm_node_t *) else_node)); - add_trace_branch_coverage(iseq, cond_seq, &branch_location, branch_location.beg_pos.column, branch_id, "else", branches); + if (exceptions->size > 0) { + for (size_t index = 0; index < exceptions->size; index++) { + PUSH_GETLOCAL(ret, *location, LVAR_ERRINFO, 0); + PM_COMPILE(exceptions->nodes[index]); + int checkmatch_flags = VM_CHECKMATCH_TYPE_RESCUE; + if (PM_NODE_TYPE_P(exceptions->nodes[index], PM_SPLAT_NODE)) { + checkmatch_flags |= VM_CHECKMATCH_ARRAY; } + PUSH_INSN1(ret, *location, checkmatch, INT2FIX(checkmatch_flags)); + PUSH_INSNL(ret, *location, branchif, exception_match_label); + } + } + else { + PUSH_GETLOCAL(ret, *location, LVAR_ERRINFO, 0); + PUSH_INSN1(ret, *location, putobject, rb_eStandardError); + PUSH_INSN1(ret, *location, checkmatch, INT2FIX(VM_CHECKMATCH_TYPE_RESCUE)); + PUSH_INSNL(ret, *location, branchif, exception_match_label); + } + + // If none of the exceptions that we are matching against matched, then + // we'll jump straight to the rescue_end_label label. + PUSH_INSNL(ret, *location, jump, rescue_end_label); + + // Here we have the exception_match_label, which is where the + // control-flow goes in the case that one of the exceptions matched. + // Here we will compile the instructions to handle the exception. + PUSH_LABEL(ret, exception_match_label); + PUSH_TRACE(ret, RUBY_EVENT_RESCUE); + + // If we have a reference to the exception, then we'll compile the write + // into the instruction sequence. This can look quite different + // depending on the kind of write being performed. + if (node->reference) { + DECL_ANCHOR(writes); + DECL_ANCHOR(cleanup); + + pm_compile_target_node(iseq, node->reference, ret, writes, cleanup, scope_node, NULL); + PUSH_GETLOCAL(ret, *location, LVAR_ERRINFO, 0); + + PUSH_SEQ(ret, writes); + PUSH_SEQ(ret, cleanup); + } + + // If we have statements to execute, we'll compile them here. Otherwise + // we'll push nil onto the stack. + if (node->statements != NULL) { + // We'll temporarily remove the end_label location from the iseq + // when compiling the statements so that next/redo statements + // inside the body will throw to the correct place instead of + // jumping straight to the end of this iseq + LABEL *prev_end = ISEQ_COMPILE_DATA(iseq)->end_label; + ISEQ_COMPILE_DATA(iseq)->end_label = NULL; + + PM_COMPILE((const pm_node_t *) node->statements); + + // Now restore the end_label + ISEQ_COMPILE_DATA(iseq)->end_label = prev_end; + } + else { + PUSH_INSN(ret, *location, putnil); + } + + PUSH_INSN(ret, *location, leave); + + // Here we'll insert the rescue_end_label label, which is jumped to if + // none of the exceptions matched. It will cause the control-flow to + // either jump to the next rescue clause or it will fall through to the + // subsequent instruction returning the raised error. + PUSH_LABEL(ret, rescue_end_label); + if (node->subsequent != NULL) { + PM_COMPILE((const pm_node_t *) node->subsequent); + } + else { + PUSH_GETLOCAL(ret, *location, 1, 0); + } +} + +static inline void +pm_compile_return_node(rb_iseq_t *iseq, const pm_return_node_t *node, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + const pm_arguments_node_t *arguments = node->arguments; + enum rb_iseq_type type = ISEQ_BODY(iseq)->type; + LABEL *splabel = 0; + + const rb_iseq_t *parent_iseq = iseq; + enum rb_iseq_type parent_type = ISEQ_BODY(parent_iseq)->type; + while (parent_type == ISEQ_TYPE_RESCUE || parent_type == ISEQ_TYPE_ENSURE) { + if (!(parent_iseq = ISEQ_BODY(parent_iseq)->parent_iseq)) break; + parent_type = ISEQ_BODY(parent_iseq)->type; + } + + switch (parent_type) { + case ISEQ_TYPE_TOP: + case ISEQ_TYPE_MAIN: + if (arguments) { + rb_warn("argument of top-level return is ignored"); + } + if (parent_iseq == iseq) { + type = ISEQ_TYPE_METHOD; + } + break; + default: + break; + } + + if (type == ISEQ_TYPE_METHOD) { + splabel = NEW_LABEL(0); + PUSH_LABEL(ret, splabel); + PUSH_ADJUST(ret, *location, 0); + } + + if (arguments != NULL) { + PM_COMPILE_NOT_POPPED((const pm_node_t *) arguments); + } + else { + PUSH_INSN(ret, *location, putnil); + } + + if (type == ISEQ_TYPE_METHOD && can_add_ensure_iseq(iseq)) { + pm_add_ensure_iseq(ret, iseq, 1, scope_node); + PUSH_TRACE(ret, RUBY_EVENT_RETURN); + PUSH_INSN(ret, *location, leave); + PUSH_ADJUST_RESTORE(ret, splabel); + if (!popped) PUSH_INSN(ret, *location, putnil); + } + else { + PUSH_INSN1(ret, *location, throw, INT2FIX(TAG_RETURN)); + if (popped) PUSH_INSN(ret, *location, pop); + } +} + +static inline void +pm_compile_super_node(rb_iseq_t *iseq, const pm_super_node_t *node, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + DECL_ANCHOR(args); + + LABEL *retry_label = NEW_LABEL(location->line); + LABEL *retry_end_l = NEW_LABEL(location->line); + + const rb_iseq_t *previous_block = ISEQ_COMPILE_DATA(iseq)->current_block; + const rb_iseq_t *current_block; + ISEQ_COMPILE_DATA(iseq)->current_block = current_block = NULL; + + PUSH_LABEL(ret, retry_label); + PUSH_INSN(ret, *location, putself); + + int flags = 0; + struct rb_callinfo_kwarg *keywords = NULL; + int argc = pm_setup_args(node->arguments, node->block, &flags, &keywords, iseq, ret, scope_node, location); + bool is_forwardable = (node->arguments != NULL) && PM_NODE_FLAG_P(node->arguments, PM_ARGUMENTS_NODE_FLAGS_CONTAINS_FORWARDING); + flags |= VM_CALL_SUPER | VM_CALL_FCALL; + + if (node->block && PM_NODE_TYPE_P(node->block, PM_BLOCK_NODE)) { + pm_scope_node_t next_scope_node; + pm_scope_node_init(node->block, &next_scope_node, scope_node); + + ISEQ_COMPILE_DATA(iseq)->current_block = current_block = NEW_CHILD_ISEQ(&next_scope_node, make_name_for_block(iseq), ISEQ_TYPE_BLOCK, location->line); + pm_scope_node_destroy(&next_scope_node); + } + + if (!node->block) { + iseq_set_use_block(ISEQ_BODY(iseq)->local_iseq); + } + + if ((flags & VM_CALL_ARGS_BLOCKARG) && (flags & VM_CALL_KW_SPLAT) && !(flags & VM_CALL_KW_SPLAT_MUT)) { + PUSH_INSN(args, *location, splatkw); + } + + PUSH_SEQ(ret, args); + if (is_forwardable && ISEQ_BODY(ISEQ_BODY(iseq)->local_iseq)->param.flags.forwardable) { + flags |= VM_CALL_FORWARDING; + + { + const struct rb_callinfo *callinfo = new_callinfo(iseq, 0, argc, flags, keywords, current_block != NULL); + PUSH_INSN2(ret, *location, invokesuperforward, callinfo, current_block); + } + } + else { + { + const struct rb_callinfo *callinfo = new_callinfo(iseq, 0, argc, flags, keywords, current_block != NULL); + PUSH_INSN2(ret, *location, invokesuper, callinfo, current_block); + } + + } + + pm_compile_retry_end_label(iseq, ret, retry_end_l); + + if (popped) PUSH_INSN(ret, *location, pop); + ISEQ_COMPILE_DATA(iseq)->current_block = previous_block; + PUSH_CATCH_ENTRY(CATCH_TYPE_BREAK, retry_label, retry_end_l, current_block, retry_end_l); +} + +static inline void +pm_compile_yield_node(rb_iseq_t *iseq, const pm_yield_node_t *node, const pm_node_location_t *location, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + switch (ISEQ_BODY(ISEQ_BODY(iseq)->local_iseq)->type) { + case ISEQ_TYPE_TOP: + case ISEQ_TYPE_MAIN: + case ISEQ_TYPE_CLASS: + COMPILE_ERROR(iseq, location->line, "Invalid yield"); + return; + default: /* valid */; + } + + int argc = 0; + int flags = 0; + struct rb_callinfo_kwarg *keywords = NULL; + + if (node->arguments) { + argc = pm_setup_args(node->arguments, NULL, &flags, &keywords, iseq, ret, scope_node, location); + } + + const struct rb_callinfo *callinfo = new_callinfo(iseq, 0, argc, flags, keywords, FALSE); + PUSH_INSN1(ret, *location, invokeblock, callinfo); + + iseq_set_use_block(ISEQ_BODY(iseq)->local_iseq); + if (popped) PUSH_INSN(ret, *location, pop); + + int level = 0; + for (const rb_iseq_t *tmp_iseq = iseq; tmp_iseq != ISEQ_BODY(iseq)->local_iseq; level++) { + tmp_iseq = ISEQ_BODY(tmp_iseq)->parent_iseq; + } + + if (level > 0) access_outer_variables(iseq, level, rb_intern("yield"), true); +} + +/** + * Compiles a prism node into instruction sequences. + * + * iseq - The current instruction sequence object (used for locals) + * node - The prism node to compile + * ret - The linked list of instructions to append instructions onto + * popped - True if compiling something with no side effects, so instructions don't + * need to be added + * scope_node - Stores parser and local information + */ +static void +pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, bool popped, pm_scope_node_t *scope_node) +{ + const pm_node_location_t location = PM_NODE_START_LOCATION(node); + int lineno = (int) location.line; + + if (PM_NODE_TYPE_P(node, PM_BEGIN_NODE) && (((const pm_begin_node_t *) node)->statements == NULL) && (((const pm_begin_node_t *) node)->rescue_clause != NULL)) { + // If this node is a begin node and it has empty statements and also + // has a rescue clause, then the other parser considers it as + // starting on the same line as the rescue, as opposed to the + // location of the begin keyword. We replicate that behavior here. + lineno = (int) PM_NODE_START_LINE_COLUMN(((const pm_begin_node_t *) node)->rescue_clause).line; + } + + if (PM_NODE_FLAG_P(node, PM_NODE_FLAG_NEWLINE) && ISEQ_COMPILE_DATA(iseq)->last_line != lineno) { + // If this node has the newline flag set and it is on a new line + // from the previous nodes that have been compiled for this ISEQ, + // then we need to emit a newline event. + int event = RUBY_EVENT_LINE; + + ISEQ_COMPILE_DATA(iseq)->last_line = lineno; + if (lineno > 0 && ISEQ_COVERAGE(iseq) && ISEQ_LINE_COVERAGE(iseq)) { + event |= RUBY_EVENT_COVERAGE_LINE; + } + PUSH_TRACE(ret, event); + } + + switch (PM_NODE_TYPE(node)) { + case PM_ALIAS_GLOBAL_VARIABLE_NODE: + // alias $foo $bar + // ^^^^^^^^^^^^^^^ + pm_compile_alias_global_variable_node(iseq, (const pm_alias_global_variable_node_t *) node, &location, ret, popped, scope_node); + return; + case PM_ALIAS_METHOD_NODE: + // alias foo bar + // ^^^^^^^^^^^^^ + pm_compile_alias_method_node(iseq, (const pm_alias_method_node_t *) node, &location, ret, popped, scope_node); + return; + case PM_AND_NODE: + // a and b + // ^^^^^^^ + pm_compile_and_node(iseq, (const pm_and_node_t *) node, &location, ret, popped, scope_node); + return; + case PM_ARGUMENTS_NODE: { + // break foo + // ^^^ + // + // These are ArgumentsNodes that are not compiled directly by their + // parent call nodes, used in the cases of NextNodes, ReturnNodes, and + // BreakNodes. They can create an array like ArrayNode. + const pm_arguments_node_t *cast = (const pm_arguments_node_t *) node; + const pm_node_list_t *elements = &cast->arguments; - PM_COMPILE_INTO_ANCHOR(cond_seq, (const pm_node_t *) else_node); - PUSH_INSNL(cond_seq, location, jump, end_label); - PUSH_INSN(cond_seq, location, putnil); - if (popped) PUSH_INSN(cond_seq, location, putnil); + if (elements->size == 1) { + // If we are only returning a single element through one of the jump + // nodes, then we will only compile that node directly. + PM_COMPILE(elements->nodes[0]); } else { - // Otherwise, if we do not have an `else` clause, we will compile in - // the code to handle raising an appropriate error. - PUSH_LABEL(cond_seq, else_label); + pm_compile_array_node(iseq, (const pm_node_t *) cast, elements, &location, ret, popped, scope_node); + } + return; + } + case PM_ARRAY_NODE: { + // [foo, bar, baz] + // ^^^^^^^^^^^^^^^ + const pm_array_node_t *cast = (const pm_array_node_t *) node; + pm_compile_array_node(iseq, (const pm_node_t *) cast, &cast->elements, &location, ret, popped, scope_node); + return; + } + case PM_ASSOC_NODE: { + // { foo: 1 } + // ^^^^^^ + // + // foo(bar: 1) + // ^^^^^^ + const pm_assoc_node_t *cast = (const pm_assoc_node_t *) node; - // Establish branch coverage for the implicit else clause. - add_trace_branch_coverage(iseq, cond_seq, &case_location, case_location.beg_pos.column, branch_id, "else", branches); + PM_COMPILE(cast->key); + PM_COMPILE(cast->value); - if (in_single_pattern) { - pm_compile_pattern_error_handler(iseq, scope_node, node, cond_seq, end_label, popped); - } - else { - PUSH_INSN1(cond_seq, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); - PUSH_INSN1(cond_seq, location, putobject, rb_eNoMatchingPatternError); - PUSH_INSN1(cond_seq, location, topn, INT2FIX(2)); - PUSH_SEND(cond_seq, location, id_core_raise, INT2FIX(2)); + return; + } + case PM_ASSOC_SPLAT_NODE: { + // { **foo } + // ^^^^^ + // + // def foo(**); bar(**); end + // ^^ + const pm_assoc_splat_node_t *cast = (const pm_assoc_splat_node_t *) node; - PUSH_INSN1(cond_seq, location, adjuststack, INT2FIX(3)); - if (!popped) PUSH_INSN(cond_seq, location, putnil); - PUSH_INSNL(cond_seq, location, jump, end_label); - PUSH_INSN1(cond_seq, location, dupn, INT2FIX(1)); - if (popped) PUSH_INSN(cond_seq, location, putnil); - } + if (cast->value != NULL) { + PM_COMPILE(cast->value); + } + else if (!popped) { + pm_local_index_t index = pm_lookup_local_index(iseq, scope_node, PM_CONSTANT_POW, 0); + PUSH_GETLOCAL(ret, location, index.index, index.level); } - // At the end of all of this compilation, we will add the code for the - // conditions first, then the various bodies, then mark the end of the - // entire sequence with the end label. - PUSH_SEQ(ret, cond_seq); - PUSH_SEQ(ret, body_seq); - PUSH_LABEL(ret, end_label); + return; + } + case PM_BACK_REFERENCE_READ_NODE: { + // $+ + // ^^ + if (!popped) { + const pm_back_reference_read_node_t *cast = (const pm_back_reference_read_node_t *) node; + VALUE backref = pm_compile_back_reference_ref(scope_node, cast); + + PUSH_INSN2(ret, location, getspecial, INT2FIX(1), backref); + } + return; + } + case PM_BEGIN_NODE: { + // begin end + // ^^^^^^^^^ + const pm_begin_node_t *cast = (const pm_begin_node_t *) node; + + if (cast->ensure_clause) { + // Compiling the ensure clause will compile the rescue clause (if + // there is one), which will compile the begin statements. + pm_compile_ensure(iseq, cast, &location, ret, popped, scope_node); + } + else if (cast->rescue_clause) { + // Compiling rescue will compile begin statements (if applicable). + pm_compile_rescue(iseq, cast, &location, ret, popped, scope_node); + } + else { + // If there is neither ensure or rescue, the just compile the + // statements. + if (cast->statements != NULL) { + PM_COMPILE((const pm_node_t *) cast->statements); + } + else if (!popped) { + PUSH_SYNTHETIC_PUTNIL(ret, iseq); + } + } + return; + } + case PM_BLOCK_ARGUMENT_NODE: { + // foo(&bar) + // ^^^^ + const pm_block_argument_node_t *cast = (const pm_block_argument_node_t *) node; + if (cast->expression != NULL) { + PM_COMPILE(cast->expression); + } + else { + // If there's no expression, this must be block forwarding. + pm_local_index_t local_index = pm_lookup_local_index(iseq, scope_node, PM_CONSTANT_AND, 0); + PUSH_INSN2(ret, location, getblockparamproxy, INT2FIX(local_index.index + VM_ENV_DATA_SIZE - 1), INT2FIX(local_index.level)); + } + return; + } + case PM_BREAK_NODE: + // break + // ^^^^^ + // + // break foo + // ^^^^^^^^^ + pm_compile_break_node(iseq, (const pm_break_node_t *) node, &location, ret, popped, scope_node); + return; + case PM_CALL_NODE: + // foo + // ^^^ + // + // foo.bar + // ^^^^^^^ + // + // foo.bar() {} + // ^^^^^^^^^^^^ + pm_compile_call_node(iseq, (const pm_call_node_t *) node, ret, popped, scope_node); + return; + case PM_CALL_AND_WRITE_NODE: { + // foo.bar &&= baz + // ^^^^^^^^^^^^^^^ + const pm_call_and_write_node_t *cast = (const pm_call_and_write_node_t *) node; + pm_compile_call_and_or_write_node(iseq, true, cast->receiver, cast->value, cast->write_name, cast->read_name, PM_NODE_FLAG_P(cast, PM_CALL_NODE_FLAGS_SAFE_NAVIGATION), &location, ret, popped, scope_node); + return; + } + case PM_CALL_OR_WRITE_NODE: { + // foo.bar ||= baz + // ^^^^^^^^^^^^^^^ + const pm_call_or_write_node_t *cast = (const pm_call_or_write_node_t *) node; + pm_compile_call_and_or_write_node(iseq, false, cast->receiver, cast->value, cast->write_name, cast->read_name, PM_NODE_FLAG_P(cast, PM_CALL_NODE_FLAGS_SAFE_NAVIGATION), &location, ret, popped, scope_node); return; } + case PM_CALL_OPERATOR_WRITE_NODE: + // foo.bar += baz + // ^^^^^^^^^^^^^^^ + // + // Call operator writes occur when you have a call node on the left-hand + // side of a write operator that is not `=`. As an example, + // `foo.bar *= 1`. This breaks down to caching the receiver on the + // stack and then performing three method calls, one to read the value, + // one to compute the result, and one to write the result back to the + // receiver. + pm_compile_call_operator_write_node(iseq, (const pm_call_operator_write_node_t *) node, &location, ret, popped, scope_node); + return; + case PM_CASE_NODE: + // case foo; when bar; end + // ^^^^^^^^^^^^^^^^^^^^^^^ + pm_compile_case_node(iseq, (const pm_case_node_t *) node, &location, ret, popped, scope_node); + return; + case PM_CASE_MATCH_NODE: + // case foo; in bar; end + // ^^^^^^^^^^^^^^^^^^^^^ + // + // If you use the `case` keyword to create a case match node, it will + // match against all of the `in` clauses until it finds one that + // matches. If it doesn't find one, it can optionally fall back to an + // `else` clause. If none is present and a match wasn't found, it will + // raise an appropriate error. + pm_compile_case_match_node(iseq, (const pm_case_match_node_t *) node, &location, ret, popped, scope_node); + return; case PM_CLASS_NODE: { // class Foo; end // ^^^^^^^^^^^^^^ @@ -6579,7 +8949,10 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, PUSH_INSN(ret, location, putnil); } - PUSH_INSN3(ret, location, defineclass, ID2SYM(class_id), class_iseq, INT2FIX(flags)); + { + VALUE operand = ID2SYM(class_id); + PUSH_INSN3(ret, location, defineclass, operand, class_iseq, INT2FIX(flags)); + } RB_OBJ_WRITTEN(iseq, Qundef, (VALUE)class_iseq); if (popped) PUSH_INSN(ret, location, pop); @@ -6686,14 +9059,12 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, if (ISEQ_COMPILE_DATA(iseq)->option->inline_const_cache && ((parts = pm_constant_path_parts(node, scope_node)) != Qnil)) { ISEQ_BODY(iseq)->ic_size++; + RB_OBJ_SET_SHAREABLE(parts); PUSH_INSN1(ret, location, opt_getconstant_path, parts); } else { DECL_ANCHOR(prefix); - INIT_ANCHOR(prefix); - DECL_ANCHOR(body); - INIT_ANCHOR(body); pm_compile_constant_path(iseq, node, prefix, body, popped, scope_node); if (LIST_INSN_SIZE_ZERO(prefix)) { @@ -6847,16 +9218,7 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, const pm_ensure_node_t *cast = (const pm_ensure_node_t *) node; if (cast->statements != NULL) { - LABEL *start = NEW_LABEL(location.line); - LABEL *end = NEW_LABEL(location.line); - PUSH_LABEL(ret, start); - - LABEL *prev_end_label = ISEQ_COMPILE_DATA(iseq)->end_label; - ISEQ_COMPILE_DATA(iseq)->end_label = end; - PM_COMPILE((const pm_node_t *) cast->statements); - ISEQ_COMPILE_DATA(iseq)->end_label = prev_end_label; - PUSH_LABEL(ret, end); } return; @@ -6899,7 +9261,8 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, // 1.0 // ^^^ if (!popped) { - PUSH_INSN1(ret, location, putobject, parse_float((const pm_float_node_t *) node)); + VALUE operand = parse_float((const pm_float_node_t *) node); + PUSH_INSN1(ret, location, putobject, operand); } return; } @@ -6936,159 +9299,17 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, PUSH_CATCH_ENTRY(CATCH_TYPE_BREAK, retry_label, retry_end_l, child_iseq, retry_end_l); return; } - case PM_FORWARDING_ARGUMENTS_NODE: { + case PM_FORWARDING_ARGUMENTS_NODE: rb_bug("Cannot compile a ForwardingArgumentsNode directly\n"); return; - } - case PM_FORWARDING_SUPER_NODE: { + case PM_FORWARDING_SUPER_NODE: // super // ^^^^^ // // super {} // ^^^^^^^^ - const pm_forwarding_super_node_t *cast = (const pm_forwarding_super_node_t *) node; - const rb_iseq_t *block = NULL; - - const rb_iseq_t *previous_block = NULL; - LABEL *retry_label = NULL; - LABEL *retry_end_l = NULL; - - if (cast->block != NULL) { - previous_block = ISEQ_COMPILE_DATA(iseq)->current_block; - ISEQ_COMPILE_DATA(iseq)->current_block = NULL; - - retry_label = NEW_LABEL(location.line); - retry_end_l = NEW_LABEL(location.line); - - PUSH_LABEL(ret, retry_label); - } - - PUSH_INSN(ret, location, putself); - int flag = VM_CALL_ZSUPER | VM_CALL_SUPER | VM_CALL_FCALL; - - if (cast->block != NULL) { - pm_scope_node_t next_scope_node; - pm_scope_node_init((const pm_node_t *) cast->block, &next_scope_node, scope_node); - - ISEQ_COMPILE_DATA(iseq)->current_block = block = NEW_CHILD_ISEQ(&next_scope_node, make_name_for_block(iseq), ISEQ_TYPE_BLOCK, location.line); - pm_scope_node_destroy(&next_scope_node); - RB_OBJ_WRITTEN(iseq, Qundef, (VALUE) block); - } - - DECL_ANCHOR(args); - INIT_ANCHOR(args); - - struct rb_iseq_constant_body *const body = ISEQ_BODY(iseq); - const rb_iseq_t *local_iseq = body->local_iseq; - const struct rb_iseq_constant_body *const local_body = ISEQ_BODY(local_iseq); - - int argc = 0; - int depth = get_lvar_level(iseq); - - if (ISEQ_BODY(ISEQ_BODY(iseq)->local_iseq)->param.flags.forwardable) { - flag |= VM_CALL_FORWARDING; - pm_local_index_t mult_local = pm_lookup_local_index(iseq, scope_node, PM_CONSTANT_DOT3, 0); - PUSH_GETLOCAL(ret, location, mult_local.index, mult_local.level); - PUSH_INSN2(ret, location, invokesuperforward, new_callinfo(iseq, 0, 0, flag, NULL, block != NULL), block); - if (popped) PUSH_INSN(ret, location, pop); - return; - } - - if (local_body->param.flags.has_lead) { - /* required arguments */ - for (int i = 0; i < local_body->param.lead_num; i++) { - int idx = local_body->local_table_size - i; - PUSH_GETLOCAL(args, location, idx, depth); - } - argc += local_body->param.lead_num; - } - - if (local_body->param.flags.has_opt) { - /* optional arguments */ - for (int j = 0; j < local_body->param.opt_num; j++) { - int idx = local_body->local_table_size - (argc + j); - PUSH_GETLOCAL(args, location, idx, depth); - } - argc += local_body->param.opt_num; - } - - if (local_body->param.flags.has_rest) { - /* rest argument */ - int idx = local_body->local_table_size - local_body->param.rest_start; - PUSH_GETLOCAL(args, location, idx, depth); - PUSH_INSN1(args, location, splatarray, Qfalse); - - argc = local_body->param.rest_start + 1; - flag |= VM_CALL_ARGS_SPLAT; - } - - if (local_body->param.flags.has_post) { - /* post arguments */ - int post_len = local_body->param.post_num; - int post_start = local_body->param.post_start; - - int j = 0; - for (; j < post_len; j++) { - int idx = local_body->local_table_size - (post_start + j); - PUSH_GETLOCAL(args, location, idx, depth); - } - - if (local_body->param.flags.has_rest) { - // argc remains unchanged from rest branch - PUSH_INSN1(args, location, newarray, INT2FIX(j)); - PUSH_INSN(args, location, concatarray); - } - else { - argc = post_len + post_start; - } - } - - const struct rb_iseq_param_keyword *const local_keyword = local_body->param.keyword; - if (local_body->param.flags.has_kw) { - int local_size = local_body->local_table_size; - argc++; - - PUSH_INSN1(args, location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); - - if (local_body->param.flags.has_kwrest) { - int idx = local_body->local_table_size - local_keyword->rest_start; - PUSH_GETLOCAL(args, location, idx, depth); - RUBY_ASSERT(local_keyword->num > 0); - PUSH_SEND(args, location, rb_intern("dup"), INT2FIX(0)); - } - else { - PUSH_INSN1(args, location, newhash, INT2FIX(0)); - } - int i = 0; - for (; i < local_keyword->num; ++i) { - ID id = local_keyword->table[i]; - int idx = local_size - get_local_var_idx(local_iseq, id); - PUSH_INSN1(args, location, putobject, ID2SYM(id)); - PUSH_GETLOCAL(args, location, idx, depth); - } - - PUSH_SEND(args, location, id_core_hash_merge_ptr, INT2FIX(i * 2 + 1)); - flag |= VM_CALL_KW_SPLAT| VM_CALL_KW_SPLAT_MUT; - } - else if (local_body->param.flags.has_kwrest) { - int idx = local_body->local_table_size - local_keyword->rest_start; - PUSH_GETLOCAL(args, location, idx, depth); - argc++; - flag |= VM_CALL_KW_SPLAT; - } - - PUSH_SEQ(ret, args); - PUSH_INSN2(ret, location, invokesuper, new_callinfo(iseq, 0, argc, flag, NULL, block != NULL), block); - - if (cast->block != NULL) { - pm_compile_retry_end_label(iseq, ret, retry_end_l); - PUSH_CATCH_ENTRY(CATCH_TYPE_BREAK, retry_label, retry_end_l, block, retry_end_l); - ISEQ_COMPILE_DATA(iseq)->current_block = previous_block; - } - - if (popped) PUSH_INSN(ret, location, pop); + pm_compile_forwarding_super_node(iseq, (const pm_forwarding_super_node_t *) node, &location, ret, popped, scope_node); return; - } case PM_GLOBAL_VARIABLE_AND_WRITE_NODE: { // $foo &&= bar // ^^^^^^^^^^^^ @@ -7223,7 +9444,7 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, } } else { - pm_compile_hash_elements(iseq, node, elements, false, ret, scope_node); + pm_compile_hash_elements(iseq, node, elements, 0, Qundef, false, ret, scope_node); } } @@ -7239,14 +9460,15 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, // foo ? bar : baz // ^^^^^^^^^^^^^^^ const pm_if_node_t *cast = (const pm_if_node_t *) node; - pm_compile_conditional(iseq, &location, PM_IF_NODE, (const pm_node_t *) cast, cast->statements, cast->consequent, cast->predicate, ret, popped, scope_node); + pm_compile_conditional(iseq, &location, PM_IF_NODE, (const pm_node_t *) cast, cast->statements, cast->subsequent, cast->predicate, ret, popped, scope_node); return; } case PM_IMAGINARY_NODE: { // 1i // ^^ if (!popped) { - PUSH_INSN1(ret, location, putobject, parse_imaginary((const pm_imaginary_node_t *) node)); + VALUE operand = parse_imaginary((const pm_imaginary_node_t *) node); + PUSH_INSN1(ret, location, putobject, operand); } return; } @@ -7382,7 +9604,8 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, // 1 // ^ if (!popped) { - PUSH_INSN1(ret, location, putobject, parse_integer((const pm_integer_node_t *) node)); + VALUE operand = parse_integer((const pm_integer_node_t *) node); + PUSH_INSN1(ret, location, putobject, operand); } return; } @@ -7416,7 +9639,7 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, pm_scope_node_t next_scope_node; pm_scope_node_init(node, &next_scope_node, scope_node); - block_iseq = NEW_CHILD_ISEQ(&next_scope_node, make_name_for_block(iseq), ISEQ_TYPE_BLOCK, location.line); + block_iseq = NEW_CHILD_ISEQ(&next_scope_node, make_name_for_block(iseq), ISEQ_TYPE_PLAIN, location.line); pm_scope_node_destroy(&next_scope_node); ISEQ_COMPILE_DATA(iseq)->current_block = block_iseq; @@ -7451,16 +9674,16 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, PUSH_INSN1(ret, location, putobject, string); } else if (PM_NODE_FLAG_P(node, PM_INTERPOLATED_STRING_NODE_FLAGS_MUTABLE)) { - PUSH_INSN1(ret, location, putstring, string); + PUSH_INSN1(ret, location, dupstring, string); } else { - PUSH_INSN1(ret, location, putchilledstring, string); + PUSH_INSN1(ret, location, dupchilledstring, string); } } } else { const pm_interpolated_string_node_t *cast = (const pm_interpolated_string_node_t *) node; - int length = pm_interpolated_node_compile(iseq, &cast->parts, &location, ret, popped, scope_node, NULL, NULL); + int length = pm_interpolated_node_compile(iseq, &cast->parts, &location, ret, popped, scope_node, NULL, NULL, PM_NODE_FLAG_P(cast, PM_INTERPOLATED_STRING_NODE_FLAGS_MUTABLE), PM_NODE_FLAG_P(cast, PM_INTERPOLATED_STRING_NODE_FLAGS_FROZEN)); if (length > 1) PUSH_INSN1(ret, location, concatstrings, INT2FIX(length)); if (popped) PUSH_INSN(ret, location, pop); } @@ -7471,7 +9694,7 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, // :"foo #{bar}" // ^^^^^^^^^^^^^ const pm_interpolated_symbol_node_t *cast = (const pm_interpolated_symbol_node_t *) node; - int length = pm_interpolated_node_compile(iseq, &cast->parts, &location, ret, popped, scope_node, NULL, NULL); + int length = pm_interpolated_node_compile(iseq, &cast->parts, &location, ret, popped, scope_node, NULL, NULL, false, false); if (length > 1) { PUSH_INSN1(ret, location, concatstrings, INT2FIX(length)); @@ -7493,7 +9716,7 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, PUSH_INSN(ret, location, putself); - int length = pm_interpolated_node_compile(iseq, &cast->parts, &location, ret, false, scope_node, NULL, NULL); + int length = pm_interpolated_node_compile(iseq, &cast->parts, &location, ret, false, scope_node, NULL, NULL, false, false); if (length > 1) PUSH_INSN1(ret, location, concatstrings, INT2FIX(length)); PUSH_SEND_WITH_FLAG(ret, location, idBackquote, INT2NUM(1), INT2FIX(VM_CALL_FCALL | VM_CALL_ARGS_SIMPLE)); @@ -7505,7 +9728,19 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, // -> { it } // ^^ if (!popped) { - PUSH_GETLOCAL(ret, location, scope_node->local_table_for_iseq_size, 0); + pm_scope_node_t *current_scope_node = scope_node; + int level = 0; + + while (current_scope_node) { + if (current_scope_node->parameters && PM_NODE_TYPE_P(current_scope_node->parameters, PM_IT_PARAMETERS_NODE)) { + PUSH_GETLOCAL(ret, location, current_scope_node->local_table_for_iseq_size, level); + return; + } + + current_scope_node = current_scope_node->previous; + level++; + } + rb_bug("Local `it` does not exist"); } return; @@ -7532,7 +9767,7 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, pm_scope_node_t next_scope_node; pm_scope_node_init(node, &next_scope_node, scope_node); - int opening_lineno = pm_location_line_number(parser, &cast->opening_loc); + int opening_lineno = pm_location_line_number_cached(&cast->opening_loc, scope_node); const rb_iseq_t *block = NEW_CHILD_ISEQ(&next_scope_node, make_name_for_block(iseq), ISEQ_TYPE_BLOCK, opening_lineno); pm_scope_node_destroy(&next_scope_node); @@ -7662,7 +9897,7 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, LABEL *matched_label = NEW_LABEL(location.line); LABEL *unmatched_label = NEW_LABEL(location.line); LABEL *done_label = NEW_LABEL(location.line); - pm_compile_pattern(iseq, scope_node, cast->pattern, ret, matched_label, unmatched_label, false, false, true, 2); + pm_compile_pattern(iseq, scope_node, cast->pattern, ret, matched_label, unmatched_label, false, true, 2); // If the pattern did not match, then compile the necessary instructions // to handle pushing false onto the stack, then jump to the end. @@ -7684,7 +9919,7 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, PUSH_LABEL(ret, done_label); return; } - case PM_MATCH_REQUIRED_NODE: { + case PM_MATCH_REQUIRED_NODE: // foo => bar // ^^^^^^^^^^ // @@ -7696,52 +9931,9 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, // This is somewhat analogous to compiling a case match statement with a // single pattern. In both cases, if the pattern fails it should // immediately raise an error. - const pm_match_required_node_t *cast = (const pm_match_required_node_t *) node; - - LABEL *matched_label = NEW_LABEL(location.line); - LABEL *unmatched_label = NEW_LABEL(location.line); - LABEL *done_label = NEW_LABEL(location.line); - - // First, we're going to push a bunch of stuff onto the stack that is - // going to serve as our scratch space. - PUSH_INSN(ret, location, putnil); // key error key - PUSH_INSN(ret, location, putnil); // key error matchee - PUSH_INSN1(ret, location, putobject, Qfalse); // key error? - PUSH_INSN(ret, location, putnil); // error string - PUSH_INSN(ret, location, putnil); // deconstruct cache - - // Next we're going to compile the value expression such that it's on - // the stack. - PM_COMPILE_NOT_POPPED(cast->value); - - // Here we'll dup it so that it can be used for comparison, but also be - // used for error handling. - PUSH_INSN(ret, location, dup); - - // Next we'll compile the pattern. We indicate to the pm_compile_pattern - // function that this is the only pattern that will be matched against - // through the in_single_pattern parameter. We also indicate that the - // value to compare against is 2 slots from the top of the stack (the - // base_index parameter). - pm_compile_pattern(iseq, scope_node, cast->pattern, ret, matched_label, unmatched_label, true, false, true, 2); - - // If the pattern did not match the value, then we're going to compile - // in our error handler code. This will determine which error to raise - // and raise it. - PUSH_LABEL(ret, unmatched_label); - pm_compile_pattern_error_handler(iseq, scope_node, node, ret, done_label, popped); - - // If the pattern did match, we'll clean up the values we've pushed onto - // the stack and then push nil onto the stack if it's not popped. - PUSH_LABEL(ret, matched_label); - PUSH_INSN1(ret, location, adjuststack, INT2FIX(6)); - if (!popped) PUSH_INSN(ret, location, putnil); - PUSH_INSNL(ret, location, jump, done_label); - - PUSH_LABEL(ret, done_label); + pm_compile_match_required_node(iseq, (const pm_match_required_node_t *) node, &location, ret, popped, scope_node); return; - } - case PM_MATCH_WRITE_NODE: { + case PM_MATCH_WRITE_NODE: // /(?<foo>foo)/ =~ bar // ^^^^^^^^^^^^^^^^^^^^ // @@ -7750,81 +9942,11 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, // operator, and some value on the right. The nodes themselves simply // wrap the call with the local variable targets that will be written // when the call is executed. - const pm_match_write_node_t *cast = (const pm_match_write_node_t *) node; - LABEL *fail_label = NEW_LABEL(location.line); - LABEL *end_label = NEW_LABEL(location.line); - - // First, we'll compile the call so that all of its instructions are - // present. Then we'll compile all of the local variable targets. - PM_COMPILE_NOT_POPPED((const pm_node_t *) cast->call); - - // Now, check if the match was successful. If it was, then we'll - // continue on and assign local variables. Otherwise we'll skip over the - // assignment code. - PUSH_INSN1(ret, location, getglobal, rb_id2sym(idBACKREF)); - PUSH_INSN(ret, location, dup); - PUSH_INSNL(ret, location, branchunless, fail_label); - - // If there's only a single local variable target, we can skip some of - // the bookkeeping, so we'll put a special branch here. - size_t targets_count = cast->targets.size; - - if (targets_count == 1) { - const pm_node_t *target = cast->targets.nodes[0]; - RUBY_ASSERT(PM_NODE_TYPE_P(target, PM_LOCAL_VARIABLE_TARGET_NODE)); - - const pm_local_variable_target_node_t *local_target = (const pm_local_variable_target_node_t *) target; - pm_local_index_t index = pm_lookup_local_index(iseq, scope_node, local_target->name, local_target->depth); - - PUSH_INSN1(ret, location, putobject, rb_id2sym(pm_constant_id_lookup(scope_node, local_target->name))); - PUSH_SEND(ret, location, idAREF, INT2FIX(1)); - PUSH_LABEL(ret, fail_label); - PUSH_SETLOCAL(ret, location, index.index, index.level); - if (popped) PUSH_INSN(ret, location, pop); - return; - } - - DECL_ANCHOR(fail_anchor); - INIT_ANCHOR(fail_anchor); - - // Otherwise there is more than one local variable target, so we'll need - // to do some bookkeeping. - for (size_t targets_index = 0; targets_index < targets_count; targets_index++) { - const pm_node_t *target = cast->targets.nodes[targets_index]; - RUBY_ASSERT(PM_NODE_TYPE_P(target, PM_LOCAL_VARIABLE_TARGET_NODE)); - - const pm_local_variable_target_node_t *local_target = (const pm_local_variable_target_node_t *) target; - pm_local_index_t index = pm_lookup_local_index(iseq, scope_node, local_target->name, local_target->depth); - - if (((size_t) targets_index) < (targets_count - 1)) { - PUSH_INSN(ret, location, dup); - } - PUSH_INSN1(ret, location, putobject, rb_id2sym(pm_constant_id_lookup(scope_node, local_target->name))); - PUSH_SEND(ret, location, idAREF, INT2FIX(1)); - PUSH_SETLOCAL(ret, location, index.index, index.level); - - PUSH_INSN(fail_anchor, location, putnil); - PUSH_SETLOCAL(fail_anchor, location, index.index, index.level); - } - - // Since we matched successfully, now we'll jump to the end. - PUSH_INSNL(ret, location, jump, end_label); - - // In the case that the match failed, we'll loop through each local - // variable target and set all of them to `nil`. - PUSH_LABEL(ret, fail_label); - PUSH_INSN(ret, location, pop); - PUSH_SEQ(ret, fail_anchor); - - // Finally, we can push the end label for either case. - PUSH_LABEL(ret, end_label); - if (popped) PUSH_INSN(ret, location, pop); + pm_compile_match_write_node(iseq, (const pm_match_write_node_t *) node, &location, ret, popped, scope_node); return; - } - case PM_MISSING_NODE: { - rb_bug("A pm_missing_node_t should not exist in prism's AST."); + case PM_ERROR_RECOVERY_NODE: + rb_bug("A pm_error_recovery_node_t should not exist in prism's AST."); return; - } case PM_MODULE_NODE: { // module Foo; end // ^^^^^^^^^^^^^^^ @@ -7868,10 +9990,7 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, const pm_multi_write_node_t *cast = (const pm_multi_write_node_t *) node; DECL_ANCHOR(writes); - INIT_ANCHOR(writes); - DECL_ANCHOR(cleanup); - INIT_ANCHOR(cleanup); pm_multi_target_state_t state = { 0 }; state.position = popped ? 0 : 1; @@ -7894,96 +10013,14 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, PUSH_SEQ(ret, cleanup); return; } - case PM_NEXT_NODE: { + case PM_NEXT_NODE: // next // ^^^^ // // next foo // ^^^^^^^^ - const pm_next_node_t *cast = (const pm_next_node_t *) node; - - if (ISEQ_COMPILE_DATA(iseq)->redo_label != 0 && can_add_ensure_iseq(iseq)) { - LABEL *splabel = NEW_LABEL(0); - PUSH_LABEL(ret, splabel); - - if (cast->arguments) { - PM_COMPILE_NOT_POPPED((const pm_node_t *) cast->arguments); - } - else { - PUSH_INSN(ret, location, putnil); - } - pm_add_ensure_iseq(ret, iseq, 0, scope_node); - - PUSH_ADJUST(ret, location, ISEQ_COMPILE_DATA(iseq)->redo_label); - PUSH_INSNL(ret, location, jump, ISEQ_COMPILE_DATA(iseq)->start_label); - - PUSH_ADJUST_RESTORE(ret, splabel); - if (!popped) PUSH_INSN(ret, location, putnil); - } - else if (ISEQ_COMPILE_DATA(iseq)->end_label && can_add_ensure_iseq(iseq)) { - LABEL *splabel = NEW_LABEL(0); - - PUSH_LABEL(ret, splabel); - PUSH_ADJUST(ret, location, ISEQ_COMPILE_DATA(iseq)->start_label); - - if (cast->arguments != NULL) { - PM_COMPILE_NOT_POPPED((const pm_node_t *) cast->arguments); - } - else { - PUSH_INSN(ret, location, putnil); - } - - pm_add_ensure_iseq(ret, iseq, 0, scope_node); - PUSH_INSNL(ret, location, jump, ISEQ_COMPILE_DATA(iseq)->end_label); - PUSH_ADJUST_RESTORE(ret, splabel); - splabel->unremovable = FALSE; - - if (!popped) PUSH_INSN(ret, location, putnil); - } - else { - const rb_iseq_t *ip = iseq; - unsigned long throw_flag = 0; - - while (ip) { - if (!ISEQ_COMPILE_DATA(ip)) { - ip = 0; - break; - } - - throw_flag = VM_THROW_NO_ESCAPE_FLAG; - if (ISEQ_COMPILE_DATA(ip)->redo_label != 0) { - /* while loop */ - break; - } - else if (ISEQ_BODY(ip)->type == ISEQ_TYPE_BLOCK) { - break; - } - else if (ISEQ_BODY(ip)->type == ISEQ_TYPE_EVAL) { - COMPILE_ERROR(iseq, location.line, "Can't escape from eval with next"); - return; - } - - ip = ISEQ_BODY(ip)->parent_iseq; - } - if (ip != 0) { - if (cast->arguments) { - PM_COMPILE_NOT_POPPED((const pm_node_t *) cast->arguments); - } - else { - PUSH_INSN(ret, location, putnil); - } - - PUSH_INSN1(ret, location, throw, INT2FIX(throw_flag | TAG_NEXT)); - if (popped) PUSH_INSN(ret, location, pop); - } - else { - COMPILE_ERROR(iseq, location.line, "Invalid next"); - return; - } - } - + pm_compile_next_node(iseq, (const pm_next_node_t *) node, &location, ret, popped, scope_node); return; - } case PM_NIL_NODE: { // nil // ^^^ @@ -7993,6 +10030,12 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, return; } + case PM_NO_BLOCK_PARAMETER_NODE: { + // def foo(&nil); end + // ^^^^ + ISEQ_BODY(iseq)->param.flags.accepts_no_block = TRUE; + return; + } case PM_NO_KEYWORDS_PARAMETER_NODE: { // def foo(**nil); end // ^^^^^ @@ -8003,10 +10046,11 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, // $1 // ^^ if (!popped) { - uint32_t reference_number = ((const pm_numbered_reference_read_node_t *) node)->number; + const pm_numbered_reference_read_node_t *cast = (const pm_numbered_reference_read_node_t *) node; - if (reference_number > 0) { - PUSH_INSN2(ret, location, getspecial, INT2FIX(1), INT2FIX(reference_number << 1)); + if (cast->number != 0) { + VALUE ref = pm_compile_numbered_reference_ref(cast); + PUSH_INSN2(ret, location, getspecial, INT2FIX(1), ref); } else { PUSH_INSN(ret, location, putnil); @@ -8073,11 +10117,9 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, // anchors and then join them in the correct order into the resulting // anchor. DECL_ANCHOR(inner_pre); - INIT_ANCHOR(inner_pre); scope_node->pre_execution_anchor = inner_pre; DECL_ANCHOR(inner_body); - INIT_ANCHOR(inner_body); if (cast->statements != NULL) { const pm_node_list_t *body = &cast->statements->body; @@ -8139,23 +10181,24 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, exclude_end ); + RB_OBJ_SET_SHAREABLE(val); PUSH_INSN1(ret, location, putobject, val); } } else { - if (cast->left == NULL) { - PUSH_INSN(ret, location, putnil); - } - else { + if (cast->left != NULL) { PM_COMPILE(cast->left); } - - if (cast->right == NULL) { + else if (!popped) { PUSH_INSN(ret, location, putnil); } - else { + + if (cast->right != NULL) { PM_COMPILE(cast->right); } + else if (!popped) { + PUSH_INSN(ret, location, putnil); + } if (!popped) { PUSH_INSN1(ret, location, newrange, INT2FIX(exclude_end ? 1 : 0)); @@ -8171,66 +10214,11 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, } return; } - case PM_REDO_NODE: { + case PM_REDO_NODE: // redo // ^^^^ - if (ISEQ_COMPILE_DATA(iseq)->redo_label && can_add_ensure_iseq(iseq)) { - LABEL *splabel = NEW_LABEL(0); - - PUSH_LABEL(ret, splabel); - PUSH_ADJUST(ret, location, ISEQ_COMPILE_DATA(iseq)->redo_label); - pm_add_ensure_iseq(ret, iseq, 0, scope_node); - - PUSH_INSNL(ret, location, jump, ISEQ_COMPILE_DATA(iseq)->redo_label); - PUSH_ADJUST_RESTORE(ret, splabel); - if (!popped) PUSH_INSN(ret, location, putnil); - } - else if (ISEQ_BODY(iseq)->type != ISEQ_TYPE_EVAL && ISEQ_COMPILE_DATA(iseq)->start_label && can_add_ensure_iseq(iseq)) { - LABEL *splabel = NEW_LABEL(0); - - PUSH_LABEL(ret, splabel); - pm_add_ensure_iseq(ret, iseq, 0, scope_node); - PUSH_ADJUST(ret, location, ISEQ_COMPILE_DATA(iseq)->start_label); - - PUSH_INSNL(ret, location, jump, ISEQ_COMPILE_DATA(iseq)->start_label); - PUSH_ADJUST_RESTORE(ret, splabel); - if (!popped) PUSH_INSN(ret, location, putnil); - } - else { - const rb_iseq_t *ip = iseq; - - while (ip) { - if (!ISEQ_COMPILE_DATA(ip)) { - ip = 0; - break; - } - - if (ISEQ_COMPILE_DATA(ip)->redo_label != 0) { - break; - } - else if (ISEQ_BODY(ip)->type == ISEQ_TYPE_BLOCK) { - break; - } - else if (ISEQ_BODY(ip)->type == ISEQ_TYPE_EVAL) { - COMPILE_ERROR(iseq, location.line, "Can't escape from eval with redo"); - return; - } - - ip = ISEQ_BODY(ip)->parent_iseq; - } - - if (ip != 0) { - PUSH_INSN(ret, location, putnil); - PUSH_INSN1(ret, location, throw, INT2FIX(VM_THROW_NO_ESCAPE_FLAG | TAG_REDO)); - if (popped) PUSH_INSN(ret, location, pop); - } - else { - COMPILE_ERROR(iseq, location.line, "Invalid redo"); - return; - } - } + pm_compile_redo_node(iseq, &location, ret, popped, scope_node); return; - } case PM_REGULAR_EXPRESSION_NODE: { // /foo/ // ^^^^^ @@ -8240,106 +10228,11 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, } return; } - case PM_RESCUE_NODE: { + case PM_RESCUE_NODE: // begin; rescue; end // ^^^^^^^ - const pm_rescue_node_t *cast = (const pm_rescue_node_t *) node; - iseq_set_exception_local_table(iseq); - - // First, establish the labels that we need to be able to jump to within - // this compilation block. - LABEL *exception_match_label = NEW_LABEL(location.line); - LABEL *rescue_end_label = NEW_LABEL(location.line); - - // Next, compile each of the exceptions that we're going to be - // handling. For each one, we'll add instructions to check if the - // exception matches the raised one, and if it does then jump to the - // exception_match_label label. Otherwise it will fall through to the - // subsequent check. If there are no exceptions, we'll only check - // StandardError. - const pm_node_list_t *exceptions = &cast->exceptions; - - if (exceptions->size > 0) { - for (size_t index = 0; index < exceptions->size; index++) { - PUSH_GETLOCAL(ret, location, LVAR_ERRINFO, 0); - PM_COMPILE(exceptions->nodes[index]); - int checkmatch_flags = VM_CHECKMATCH_TYPE_RESCUE; - if (PM_NODE_TYPE_P(exceptions->nodes[index], PM_SPLAT_NODE)) { - checkmatch_flags |= VM_CHECKMATCH_ARRAY; - } - PUSH_INSN1(ret, location, checkmatch, INT2FIX(checkmatch_flags)); - PUSH_INSNL(ret, location, branchif, exception_match_label); - } - } - else { - PUSH_GETLOCAL(ret, location, LVAR_ERRINFO, 0); - PUSH_INSN1(ret, location, putobject, rb_eStandardError); - PUSH_INSN1(ret, location, checkmatch, INT2FIX(VM_CHECKMATCH_TYPE_RESCUE)); - PUSH_INSNL(ret, location, branchif, exception_match_label); - } - - // If none of the exceptions that we are matching against matched, then - // we'll jump straight to the rescue_end_label label. - PUSH_INSNL(ret, location, jump, rescue_end_label); - - // Here we have the exception_match_label, which is where the - // control-flow goes in the case that one of the exceptions matched. - // Here we will compile the instructions to handle the exception. - PUSH_LABEL(ret, exception_match_label); - PUSH_TRACE(ret, RUBY_EVENT_RESCUE); - - // If we have a reference to the exception, then we'll compile the write - // into the instruction sequence. This can look quite different - // depending on the kind of write being performed. - if (cast->reference) { - DECL_ANCHOR(writes); - INIT_ANCHOR(writes); - - DECL_ANCHOR(cleanup); - INIT_ANCHOR(cleanup); - - pm_compile_target_node(iseq, cast->reference, ret, writes, cleanup, scope_node, NULL); - PUSH_GETLOCAL(ret, location, LVAR_ERRINFO, 0); - - PUSH_SEQ(ret, writes); - PUSH_SEQ(ret, cleanup); - } - - // If we have statements to execute, we'll compile them here. Otherwise - // we'll push nil onto the stack. - if (cast->statements) { - // We'll temporarily remove the end_label location from the iseq - // when compiling the statements so that next/redo statements - // inside the body will throw to the correct place instead of - // jumping straight to the end of this iseq - LABEL *prev_end = ISEQ_COMPILE_DATA(iseq)->end_label; - ISEQ_COMPILE_DATA(iseq)->end_label = NULL; - - PM_COMPILE((const pm_node_t *) cast->statements); - - // Now restore the end_label - ISEQ_COMPILE_DATA(iseq)->end_label = prev_end; - } - else { - PUSH_INSN(ret, location, putnil); - } - - PUSH_INSN(ret, location, leave); - - // Here we'll insert the rescue_end_label label, which is jumped to if - // none of the exceptions matched. It will cause the control-flow to - // either jump to the next rescue clause or it will fall through to the - // subsequent instruction returning the raised error. - PUSH_LABEL(ret, rescue_end_label); - if (cast->consequent) { - PM_COMPILE((const pm_node_t *) cast->consequent); - } - else { - PUSH_GETLOCAL(ret, location, 1, 0); - } - + pm_compile_rescue_node(iseq, (const pm_rescue_node_t *) node, &location, ret, popped, scope_node); return; - } case PM_RESCUE_MODIFIER_NODE: { // foo rescue bar // ^^^^^^^^^^^^^^ @@ -8352,7 +10245,7 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, &rescue_scope_node, rb_str_concat(rb_str_new2("rescue in "), ISEQ_BODY(iseq)->location.label), ISEQ_TYPE_RESCUE, - pm_node_line_number(parser, cast->rescue_expression) + pm_node_line_number_cached(cast->rescue_expression, scope_node) ); pm_scope_node_destroy(&rescue_scope_node); @@ -8376,66 +10269,14 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, PUSH_CATCH_ENTRY(CATCH_TYPE_RETRY, lend, lcont, NULL, lstart); return; } - case PM_RETURN_NODE: { + case PM_RETURN_NODE: // return // ^^^^^^ // // return 1 // ^^^^^^^^ - const pm_return_node_t *cast = (const pm_return_node_t *) node; - const pm_arguments_node_t *arguments = cast->arguments; - - enum rb_iseq_type type = ISEQ_BODY(iseq)->type; - LABEL *splabel = 0; - - const rb_iseq_t *parent_iseq = iseq; - enum rb_iseq_type parent_type = ISEQ_BODY(parent_iseq)->type; - while (parent_type == ISEQ_TYPE_RESCUE || parent_type == ISEQ_TYPE_ENSURE) { - if (!(parent_iseq = ISEQ_BODY(parent_iseq)->parent_iseq)) break; - parent_type = ISEQ_BODY(parent_iseq)->type; - } - - switch (parent_type) { - case ISEQ_TYPE_TOP: - case ISEQ_TYPE_MAIN: - if (arguments) { - rb_warn("argument of top-level return is ignored"); - } - if (parent_iseq == iseq) { - type = ISEQ_TYPE_METHOD; - } - break; - default: - break; - } - - if (type == ISEQ_TYPE_METHOD) { - splabel = NEW_LABEL(0); - PUSH_LABEL(ret, splabel); - PUSH_ADJUST(ret, location, 0); - } - - if (arguments) { - PM_COMPILE_NOT_POPPED((const pm_node_t *) arguments); - } - else { - PUSH_INSN(ret, location, putnil); - } - - if (type == ISEQ_TYPE_METHOD && can_add_ensure_iseq(iseq)) { - pm_add_ensure_iseq(ret, iseq, 1, scope_node); - PUSH_TRACE(ret, RUBY_EVENT_RETURN); - PUSH_INSN(ret, location, leave); - PUSH_ADJUST_RESTORE(ret, splabel); - if (!popped) PUSH_INSN(ret, location, putnil); - } - else { - PUSH_INSN1(ret, location, throw, INT2FIX(TAG_RETURN)); - if (popped) PUSH_INSN(ret, location, pop); - } - + pm_compile_return_node(iseq, (const pm_return_node_t *) node, &location, ret, popped, scope_node); return; - } case PM_RETRY_NODE: { // retry // ^^^^^ @@ -8450,919 +10291,9 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, } return; } - case PM_SCOPE_NODE: { - pm_scope_node_t *scope_node = (pm_scope_node_t *) node; - pm_constant_id_list_t *locals = &scope_node->locals; - - pm_parameters_node_t *parameters_node = NULL; - pm_node_list_t *keywords_list = NULL; - pm_node_list_t *optionals_list = NULL; - pm_node_list_t *posts_list = NULL; - pm_node_list_t *requireds_list = NULL; - pm_node_list_t *block_locals = NULL; - bool trailing_comma = false; - - struct rb_iseq_constant_body *body = ISEQ_BODY(iseq); - - if (PM_NODE_TYPE_P(scope_node->ast_node, PM_CLASS_NODE) || PM_NODE_TYPE_P(scope_node->ast_node, PM_MODULE_NODE)) { - ADD_TRACE(ret, RUBY_EVENT_CLASS); - } - - if (scope_node->parameters) { - switch (PM_NODE_TYPE(scope_node->parameters)) { - case PM_BLOCK_PARAMETERS_NODE: { - pm_block_parameters_node_t *cast = (pm_block_parameters_node_t *) scope_node->parameters; - parameters_node = cast->parameters; - block_locals = &cast->locals; - - if (parameters_node) { - if (parameters_node->rest && PM_NODE_TYPE_P(parameters_node->rest, PM_IMPLICIT_REST_NODE)) { - trailing_comma = true; - } - } - break; - } - case PM_PARAMETERS_NODE: { - parameters_node = (pm_parameters_node_t *) scope_node->parameters; - break; - } - case PM_NUMBERED_PARAMETERS_NODE: { - uint32_t maximum = ((const pm_numbered_parameters_node_t *) scope_node->parameters)->maximum; - body->param.lead_num = maximum; - body->param.flags.ambiguous_param0 = maximum == 1; - break; - } - case PM_IT_PARAMETERS_NODE: - body->param.lead_num = 1; - body->param.flags.ambiguous_param0 = true; - break; - default: - rb_bug("Unexpected node type for parameters: %s", pm_node_type_to_str(PM_NODE_TYPE(node))); - } - } - - struct rb_iseq_param_keyword *keyword = NULL; - - if (parameters_node) { - optionals_list = ¶meters_node->optionals; - requireds_list = ¶meters_node->requireds; - keywords_list = ¶meters_node->keywords; - posts_list = ¶meters_node->posts; - } - else if (scope_node->parameters && (PM_NODE_TYPE_P(scope_node->parameters, PM_NUMBERED_PARAMETERS_NODE) || PM_NODE_TYPE_P(scope_node->parameters, PM_IT_PARAMETERS_NODE))) { - body->param.opt_num = 0; - } - else { - body->param.lead_num = 0; - body->param.opt_num = 0; - } - - //********STEP 1********** - // Goal: calculate the table size for the locals, accounting for - // hidden variables and multi target nodes - size_t locals_size = locals->size; - - // Index lookup table buffer size is only the number of the locals - st_table *index_lookup_table = st_init_numtable(); - - int table_size = (int) locals_size; - - // For nodes have a hidden iteration variable. We add that to the local - // table size here. - if (PM_NODE_TYPE_P(scope_node->ast_node, PM_FOR_NODE)) table_size++; - - if (keywords_list && keywords_list->size) { - table_size++; - } - - if (requireds_list) { - for (size_t i = 0; i < requireds_list->size; i++) { - // For each MultiTargetNode, we're going to have one - // additional anonymous local not represented in the locals table - // We want to account for this in our table size - pm_node_t *required = requireds_list->nodes[i]; - if (PM_NODE_TYPE_P(required, PM_MULTI_TARGET_NODE)) { - table_size++; - } - else if (PM_NODE_TYPE_P(required, PM_REQUIRED_PARAMETER_NODE)) { - if (PM_NODE_FLAG_P(required, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { - table_size++; - } - } - } - } - - // If we have the `it` implicit local variable, we need to account for - // it in the local table size. - if (scope_node->parameters != NULL && PM_NODE_TYPE_P(scope_node->parameters, PM_IT_PARAMETERS_NODE)) { - table_size++; - } - - // Ensure there is enough room in the local table for any - // parameters that have been repeated - // ex: def underscore_parameters(_, _ = 1, _ = 2); _; end - // ^^^^^^^^^^^^ - if (optionals_list && optionals_list->size) { - for (size_t i = 0; i < optionals_list->size; i++) { - pm_node_t * node = optionals_list->nodes[i]; - if (PM_NODE_FLAG_P(node, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { - table_size++; - } - } - } - - // If we have an anonymous "rest" node, we'll need to increase the local - // table size to take it in to account. - // def m(foo, *, bar) - // ^ - if (parameters_node) { - if (parameters_node->rest) { - if (!(PM_NODE_TYPE_P(parameters_node->rest, PM_IMPLICIT_REST_NODE))) { - if (!((const pm_rest_parameter_node_t *) parameters_node->rest)->name || PM_NODE_FLAG_P(parameters_node->rest, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { - table_size++; - } - } - } - - // def foo(_, **_); _; end - // ^^^ - if (parameters_node->keyword_rest) { - // def foo(...); end - // ^^^ - // When we have a `...` as the keyword_rest, it's a forwarding_parameter_node and - // we need to leave space for 4 locals: *, **, &, ... - if (PM_NODE_TYPE_P(parameters_node->keyword_rest, PM_FORWARDING_PARAMETER_NODE)) { - // Only optimize specifically methods like this: `foo(...)` - if (requireds_list->size == 0 && optionals_list->size == 0 && keywords_list->size == 0) { - ISEQ_BODY(iseq)->param.flags.forwardable = TRUE; - table_size += 1; - } - else { - table_size += 4; - } - } - else { - const pm_keyword_rest_parameter_node_t *kw_rest = (const pm_keyword_rest_parameter_node_t *) parameters_node->keyword_rest; - - // If it's anonymous or repeated, then we need to allocate stack space - if (!kw_rest->name || PM_NODE_FLAG_P(kw_rest, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { - table_size++; - } - } - } - } - - if (posts_list) { - for (size_t i = 0; i < posts_list->size; i++) { - // For each MultiTargetNode, we're going to have one - // additional anonymous local not represented in the locals table - // We want to account for this in our table size - pm_node_t *required = posts_list->nodes[i]; - if (PM_NODE_TYPE_P(required, PM_MULTI_TARGET_NODE) || PM_NODE_FLAG_P(required, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { - table_size++; - } - } - } - - if (keywords_list && keywords_list->size) { - for (size_t i = 0; i < keywords_list->size; i++) { - pm_node_t *keyword_parameter_node = keywords_list->nodes[i]; - if (PM_NODE_FLAG_P(keyword_parameter_node, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { - table_size++; - } - } - } - - if (parameters_node && parameters_node->block) { - const pm_block_parameter_node_t *block_node = (const pm_block_parameter_node_t *) parameters_node->block; - - if (PM_NODE_FLAG_P(block_node, PM_PARAMETER_FLAGS_REPEATED_PARAMETER) || !block_node->name) { - table_size++; - } - } - - // We can create local_table_for_iseq with the correct size - VALUE idtmp = 0; - rb_ast_id_table_t *local_table_for_iseq = ALLOCV(idtmp, sizeof(rb_ast_id_table_t) + table_size * sizeof(ID)); - local_table_for_iseq->size = table_size; - - //********END OF STEP 1********** - - //********STEP 2********** - // Goal: populate iv index table as well as local table, keeping the - // layout of the local table consistent with the layout of the - // stack when calling the method - // - // Do a first pass on all of the parameters, setting their values in - // the local_table_for_iseq, _except_ for Multis who get a hidden - // variable in this step, and will get their names inserted in step 3 - - // local_index is a cursor that keeps track of the current - // index into local_table_for_iseq. The local table is actually a list, - // and the order of that list must match the order of the items pushed - // on the stack. We need to take in to account things pushed on the - // stack that _might not have a name_ (for example array destructuring). - // This index helps us know which item we're dealing with and also give - // those anonymous items temporary names (as below) - int local_index = 0; - - // Here we figure out local table indices and insert them in to the - // index lookup table and local tables. - // - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^^^^^^^^^^^^ - if (requireds_list && requireds_list->size) { - for (size_t i = 0; i < requireds_list->size; i++, local_index++) { - ID local; - - // For each MultiTargetNode, we're going to have one additional - // anonymous local not represented in the locals table. We want - // to account for this in our table size. - pm_node_t *required = requireds_list->nodes[i]; - - switch (PM_NODE_TYPE(required)) { - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^^^^^^^^^ - case PM_MULTI_TARGET_NODE: { - local = rb_make_temporary_id(local_index); - local_table_for_iseq->ids[local_index] = local; - break; - } - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^ - case PM_REQUIRED_PARAMETER_NODE: { - const pm_required_parameter_node_t *param = (const pm_required_parameter_node_t *) required; - - if (PM_NODE_FLAG_P(required, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { - ID local = pm_constant_id_lookup(scope_node, param->name); - local_table_for_iseq->ids[local_index] = local; - } - else { - pm_insert_local_index(param->name, local_index, index_lookup_table, local_table_for_iseq, scope_node); - } - - break; - } - default: { - rb_bug("Unsupported node in requireds in parameters %s", pm_node_type_to_str(PM_NODE_TYPE(node))); - } - } - } - - body->param.lead_num = (int) requireds_list->size; - body->param.flags.has_lead = true; - } - - if (scope_node->parameters != NULL && PM_NODE_TYPE_P(scope_node->parameters, PM_IT_PARAMETERS_NODE)) { - ID local = rb_make_temporary_id(local_index); - local_table_for_iseq->ids[local_index++] = local; - } - - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^^^^ - if (optionals_list && optionals_list->size) { - body->param.opt_num = (int) optionals_list->size; - body->param.flags.has_opt = true; - - for (size_t i = 0; i < optionals_list->size; i++, local_index++) { - pm_node_t * node = optionals_list->nodes[i]; - pm_constant_id_t name = ((const pm_optional_parameter_node_t *) node)->name; - - if (PM_NODE_FLAG_P(node, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { - ID local = pm_constant_id_lookup(scope_node, name); - local_table_for_iseq->ids[local_index] = local; - } - else { - pm_insert_local_index(name, local_index, index_lookup_table, local_table_for_iseq, scope_node); - } - } - } - - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^ - if (parameters_node && parameters_node->rest) { - body->param.rest_start = local_index; - - // If there's a trailing comma, we'll have an implicit rest node, - // and we don't want it to impact the rest variables on param - if (!(PM_NODE_TYPE_P(parameters_node->rest, PM_IMPLICIT_REST_NODE))) { - body->param.flags.has_rest = true; - RUBY_ASSERT(body->param.rest_start != -1); - - pm_constant_id_t name = ((const pm_rest_parameter_node_t *) parameters_node->rest)->name; - - if (name) { - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^ - if (PM_NODE_FLAG_P(parameters_node->rest, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { - ID local = pm_constant_id_lookup(scope_node, name); - local_table_for_iseq->ids[local_index] = local; - } - else { - pm_insert_local_index(name, local_index, index_lookup_table, local_table_for_iseq, scope_node); - } - } - else { - // def foo(a, (b, *c, d), e = 1, *, g, (h, *i, j), k:, l: 1, **m, &n) - // ^ - body->param.flags.anon_rest = true; - pm_insert_local_special(idMULT, local_index, index_lookup_table, local_table_for_iseq); - } - - local_index++; - } - } - - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^^^^^^^^^^^^ - if (posts_list && posts_list->size) { - body->param.post_num = (int) posts_list->size; - body->param.post_start = local_index; - body->param.flags.has_post = true; - - for (size_t i = 0; i < posts_list->size; i++, local_index++) { - ID local; - - // For each MultiTargetNode, we're going to have one additional - // anonymous local not represented in the locals table. We want - // to account for this in our table size. - const pm_node_t *post_node = posts_list->nodes[i]; - - switch (PM_NODE_TYPE(post_node)) { - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^^^^^^^^^ - case PM_MULTI_TARGET_NODE: { - local = rb_make_temporary_id(local_index); - local_table_for_iseq->ids[local_index] = local; - break; - } - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^ - case PM_REQUIRED_PARAMETER_NODE: { - const pm_required_parameter_node_t *param = (const pm_required_parameter_node_t *) post_node; - - if (PM_NODE_FLAG_P(param, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { - ID local = pm_constant_id_lookup(scope_node, param->name); - local_table_for_iseq->ids[local_index] = local; - } - else { - pm_insert_local_index(param->name, local_index, index_lookup_table, local_table_for_iseq, scope_node); - } - break; - } - default: { - rb_bug("Unsupported node in posts in parameters %s", pm_node_type_to_str(PM_NODE_TYPE(node))); - } - } - } - } - - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^^^^^^^ - // Keywords create an internal variable on the parse tree - if (keywords_list && keywords_list->size) { - body->param.keyword = keyword = ZALLOC_N(struct rb_iseq_param_keyword, 1); - keyword->num = (int) keywords_list->size; - - body->param.flags.has_kw = true; - const VALUE default_values = rb_ary_hidden_new(1); - const VALUE complex_mark = rb_str_tmp_new(0); - - ID *ids = xcalloc(keywords_list->size, sizeof(ID)); - - size_t kw_index = 0; - - for (size_t i = 0; i < keywords_list->size; i++) { - pm_node_t *keyword_parameter_node = keywords_list->nodes[i]; - pm_constant_id_t name; - - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^ - if (PM_NODE_TYPE_P(keyword_parameter_node, PM_REQUIRED_KEYWORD_PARAMETER_NODE)) { - name = ((const pm_required_keyword_parameter_node_t *) keyword_parameter_node)->name; - keyword->required_num++; - ID local = pm_constant_id_lookup(scope_node, name); - - if (PM_NODE_FLAG_P(keyword_parameter_node, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { - local_table_for_iseq->ids[local_index] = local; - } - else { - pm_insert_local_index(name, local_index, index_lookup_table, local_table_for_iseq, scope_node); - } - local_index++; - ids[kw_index++] = local; - } - } - - for (size_t i = 0; i < keywords_list->size; i++) { - pm_node_t *keyword_parameter_node = keywords_list->nodes[i]; - pm_constant_id_t name; - - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^^^ - if (PM_NODE_TYPE_P(keyword_parameter_node, PM_OPTIONAL_KEYWORD_PARAMETER_NODE)) { - const pm_optional_keyword_parameter_node_t *cast = ((const pm_optional_keyword_parameter_node_t *) keyword_parameter_node); - - pm_node_t *value = cast->value; - name = cast->name; - - if (PM_NODE_FLAG_P(value, PM_NODE_FLAG_STATIC_LITERAL) && !(PM_NODE_TYPE_P(value, PM_ARRAY_NODE) || PM_NODE_TYPE_P(value, PM_HASH_NODE) || PM_NODE_TYPE_P(value, PM_RANGE_NODE))) { - rb_ary_push(default_values, pm_static_literal_value(iseq, value, scope_node)); - } - else { - rb_ary_push(default_values, complex_mark); - } - - ID local = pm_constant_id_lookup(scope_node, name); - if (PM_NODE_FLAG_P(keyword_parameter_node, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { - local_table_for_iseq->ids[local_index] = local; - } - else { - pm_insert_local_index(name, local_index, index_lookup_table, local_table_for_iseq, scope_node); - } - ids[kw_index++] = local; - local_index++; - } - - } - - keyword->bits_start = local_index; - keyword->table = ids; - - VALUE *dvs = ALLOC_N(VALUE, RARRAY_LEN(default_values)); - - for (int i = 0; i < RARRAY_LEN(default_values); i++) { - VALUE dv = RARRAY_AREF(default_values, i); - if (dv == complex_mark) dv = Qundef; - if (!SPECIAL_CONST_P(dv)) { - RB_OBJ_WRITTEN(iseq, Qundef, dv); - } - dvs[i] = dv; - } - - keyword->default_values = dvs; - - // Hidden local for keyword arguments - ID local = rb_make_temporary_id(local_index); - local_table_for_iseq->ids[local_index] = local; - local_index++; - } - - if (body->type == ISEQ_TYPE_BLOCK && local_index == 1 && requireds_list && requireds_list->size == 1 && !trailing_comma) { - body->param.flags.ambiguous_param0 = true; - } - - if (parameters_node) { - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^^ - if (parameters_node->keyword_rest) { - switch (PM_NODE_TYPE(parameters_node->keyword_rest)) { - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **nil, &n) - // ^^^^^ - case PM_NO_KEYWORDS_PARAMETER_NODE: { - body->param.flags.accepts_no_kwarg = true; - break; - } - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^^ - case PM_KEYWORD_REST_PARAMETER_NODE: { - const pm_keyword_rest_parameter_node_t *kw_rest_node = (const pm_keyword_rest_parameter_node_t *) parameters_node->keyword_rest; - if (!body->param.flags.has_kw) { - body->param.keyword = keyword = ZALLOC_N(struct rb_iseq_param_keyword, 1); - } - - keyword->rest_start = local_index; - body->param.flags.has_kwrest = true; - - pm_constant_id_t constant_id = kw_rest_node->name; - if (constant_id) { - if (PM_NODE_FLAG_P(kw_rest_node, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { - ID local = pm_constant_id_lookup(scope_node, constant_id); - local_table_for_iseq->ids[local_index] = local; - } - else { - pm_insert_local_index(constant_id, local_index, index_lookup_table, local_table_for_iseq, scope_node); - } - } - else { - body->param.flags.anon_kwrest = true; - pm_insert_local_special(idPow, local_index, index_lookup_table, local_table_for_iseq); - } - - local_index++; - break; - } - // def foo(...) - // ^^^ - case PM_FORWARDING_PARAMETER_NODE: { - if (!ISEQ_BODY(iseq)->param.flags.forwardable) { - // Add the anonymous * - body->param.rest_start = local_index; - body->param.flags.has_rest = true; - body->param.flags.anon_rest = true; - pm_insert_local_special(idMULT, local_index++, index_lookup_table, local_table_for_iseq); - - // Add the anonymous ** - RUBY_ASSERT(!body->param.flags.has_kw); - body->param.flags.has_kw = false; - body->param.flags.has_kwrest = true; - body->param.flags.anon_kwrest = true; - body->param.keyword = keyword = ZALLOC_N(struct rb_iseq_param_keyword, 1); - keyword->rest_start = local_index; - pm_insert_local_special(idPow, local_index++, index_lookup_table, local_table_for_iseq); - - // Add the anonymous & - body->param.block_start = local_index; - body->param.flags.has_block = true; - pm_insert_local_special(idAnd, local_index++, index_lookup_table, local_table_for_iseq); - } - - // Add the ... - pm_insert_local_special(idDot3, local_index++, index_lookup_table, local_table_for_iseq); - break; - } - default: { - rb_bug("node type %s not expected as keyword_rest", pm_node_type_to_str(PM_NODE_TYPE(parameters_node->keyword_rest))); - } - } - } - - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^ - if (parameters_node->block) { - body->param.block_start = local_index; - body->param.flags.has_block = true; - - pm_constant_id_t name = ((const pm_block_parameter_node_t *) parameters_node->block)->name; - - if (name) { - if (PM_NODE_FLAG_P(parameters_node->block, PM_PARAMETER_FLAGS_REPEATED_PARAMETER)) { - ID local = pm_constant_id_lookup(scope_node, name); - local_table_for_iseq->ids[local_index] = local; - } - else { - pm_insert_local_index(name, local_index, index_lookup_table, local_table_for_iseq, scope_node); - } - } - else { - pm_insert_local_special(idAnd, local_index, index_lookup_table, local_table_for_iseq); - } - - local_index++; - } - } - - //********END OF STEP 2********** - // The local table is now consistent with expected - // stack layout - - // If there's only one required element in the parameters - // CRuby needs to recognize it as an ambiguous parameter - - //********STEP 3********** - // Goal: fill in the names of the parameters in MultiTargetNodes - // - // Go through requireds again to set the multis - - if (requireds_list && requireds_list->size) { - for (size_t i = 0; i < requireds_list->size; i++) { - // For each MultiTargetNode, we're going to have one - // additional anonymous local not represented in the locals table - // We want to account for this in our table size - const pm_node_t *required = requireds_list->nodes[i]; - - if (PM_NODE_TYPE_P(required, PM_MULTI_TARGET_NODE)) { - local_index = pm_compile_destructured_param_locals((const pm_multi_target_node_t *) required, index_lookup_table, local_table_for_iseq, scope_node, local_index); - } - } - } - - // Go through posts again to set the multis - if (posts_list && posts_list->size) { - for (size_t i = 0; i < posts_list->size; i++) { - // For each MultiTargetNode, we're going to have one - // additional anonymous local not represented in the locals table - // We want to account for this in our table size - const pm_node_t *post = posts_list->nodes[i]; - - if (PM_NODE_TYPE_P(post, PM_MULTI_TARGET_NODE)) { - local_index = pm_compile_destructured_param_locals((const pm_multi_target_node_t *) post, index_lookup_table, local_table_for_iseq, scope_node, local_index); - } - } - } - - // Set any anonymous locals for the for node - if (PM_NODE_TYPE_P(scope_node->ast_node, PM_FOR_NODE)) { - if (PM_NODE_TYPE_P(((const pm_for_node_t *) scope_node->ast_node)->index, PM_LOCAL_VARIABLE_TARGET_NODE)) { - body->param.lead_num++; - } - else { - body->param.rest_start = local_index; - body->param.flags.has_rest = true; - } - - ID local = rb_make_temporary_id(local_index); - local_table_for_iseq->ids[local_index] = local; - local_index++; - } - - // Fill in any NumberedParameters, if they exist - if (scope_node->parameters && PM_NODE_TYPE_P(scope_node->parameters, PM_NUMBERED_PARAMETERS_NODE)) { - int maximum = ((const pm_numbered_parameters_node_t *) scope_node->parameters)->maximum; - RUBY_ASSERT(0 < maximum && maximum <= 9); - for (int i = 0; i < maximum; i++, local_index++) { - const uint8_t param_name[] = { '_', '1' + i }; - pm_constant_id_t constant_id = pm_constant_pool_find(&parser->constant_pool, param_name, 2); - RUBY_ASSERT(constant_id && "parser should fill in any gaps in numbered parameters"); - pm_insert_local_index(constant_id, local_index, index_lookup_table, local_table_for_iseq, scope_node); - } - body->param.lead_num = maximum; - body->param.flags.has_lead = true; - } - - //********END OF STEP 3********** - - //********STEP 4********** - // Goal: fill in the method body locals - // To be explicit, these are the non-parameter locals - // We fill in the block_locals, if they exist - // lambda { |x; y| y } - // ^ - if (block_locals && block_locals->size) { - for (size_t i = 0; i < block_locals->size; i++, local_index++) { - pm_constant_id_t constant_id = ((const pm_block_local_variable_node_t *) block_locals->nodes[i])->name; - pm_insert_local_index(constant_id, local_index, index_lookup_table, local_table_for_iseq, scope_node); - } - } - - // Fill in any locals we missed - if (scope_node->locals.size) { - for (size_t i = 0; i < scope_node->locals.size; i++) { - pm_constant_id_t constant_id = locals->ids[i]; - if (constant_id) { - struct pm_local_table_insert_ctx ctx; - ctx.scope_node = scope_node; - ctx.local_table_for_iseq = local_table_for_iseq; - ctx.local_index = local_index; - - st_update(index_lookup_table, (st_data_t)constant_id, pm_local_table_insert_func, (st_data_t)&ctx); - - local_index = ctx.local_index; - } - } - } - - //********END OF STEP 4********** - - // We set the index_lookup_table on the scope node so we can - // refer to the parameters correctly - if (scope_node->index_lookup_table) { - st_free_table(scope_node->index_lookup_table); - } - scope_node->index_lookup_table = index_lookup_table; - iseq_calc_param_size(iseq); - - if (ISEQ_BODY(iseq)->param.flags.forwardable) { - // We're treating `...` as a parameter so that frame - // pushing won't clobber it. - ISEQ_BODY(iseq)->param.size += 1; - } - - // FIXME: args? - iseq_set_local_table(iseq, local_table_for_iseq, 0); - scope_node->local_table_for_iseq_size = local_table_for_iseq->size; - - //********STEP 5************ - // Goal: compile anything that needed to be compiled - if (optionals_list && optionals_list->size) { - LABEL **opt_table = (LABEL **) ALLOC_N(VALUE, optionals_list->size + 1); - LABEL *label; - - // TODO: Should we make an api for NEW_LABEL where you can pass - // a pointer to the label it should fill out? We already - // have a list of labels allocated above so it seems wasteful - // to do the copies. - for (size_t i = 0; i < optionals_list->size; i++) { - label = NEW_LABEL(lineno); - opt_table[i] = label; - PUSH_LABEL(ret, label); - pm_node_t *optional_node = optionals_list->nodes[i]; - PM_COMPILE_NOT_POPPED(optional_node); - } - - // Set the last label - label = NEW_LABEL(lineno); - opt_table[optionals_list->size] = label; - PUSH_LABEL(ret, label); - - body->param.opt_table = (const VALUE *) opt_table; - } - - if (keywords_list && keywords_list->size) { - size_t optional_index = 0; - for (size_t i = 0; i < keywords_list->size; i++) { - pm_node_t *keyword_parameter_node = keywords_list->nodes[i]; - pm_constant_id_t name; - - switch (PM_NODE_TYPE(keyword_parameter_node)) { - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^^^ - case PM_OPTIONAL_KEYWORD_PARAMETER_NODE: { - const pm_optional_keyword_parameter_node_t *cast = ((const pm_optional_keyword_parameter_node_t *) keyword_parameter_node); - - pm_node_t *value = cast->value; - name = cast->name; - - if (!PM_NODE_FLAG_P(value, PM_NODE_FLAG_STATIC_LITERAL) || PM_NODE_TYPE_P(value, PM_ARRAY_NODE) || PM_NODE_TYPE_P(value, PM_HASH_NODE) || PM_NODE_TYPE_P(value, PM_RANGE_NODE)) { - LABEL *end_label = NEW_LABEL(location.line); - - pm_local_index_t index = pm_lookup_local_index(iseq, scope_node, name, 0); - int kw_bits_idx = table_size - body->param.keyword->bits_start; - PUSH_INSN2(ret, location, checkkeyword, INT2FIX(kw_bits_idx + VM_ENV_DATA_SIZE - 1), INT2FIX(optional_index)); - PUSH_INSNL(ret, location, branchif, end_label); - PM_COMPILE(value); - PUSH_SETLOCAL(ret, location, index.index, index.level); - PUSH_LABEL(ret, end_label); - } - optional_index++; - break; - } - // def foo(a, (b, *c, d), e = 1, *f, g, (h, *i, j), k:, l: 1, **m, &n) - // ^^ - case PM_REQUIRED_KEYWORD_PARAMETER_NODE: { - break; - } - default: { - rb_bug("Unexpected keyword parameter node type %s", pm_node_type_to_str(PM_NODE_TYPE(keyword_parameter_node))); - } - } - } - } - - if (requireds_list && requireds_list->size) { - for (size_t i = 0; i < requireds_list->size; i++) { - // For each MultiTargetNode, we're going to have one additional - // anonymous local not represented in the locals table. We want - // to account for this in our table size. - const pm_node_t *required = requireds_list->nodes[i]; - - if (PM_NODE_TYPE_P(required, PM_MULTI_TARGET_NODE)) { - PUSH_GETLOCAL(ret, location, table_size - (int)i, 0); - pm_compile_destructured_param_writes(iseq, (const pm_multi_target_node_t *) required, ret, scope_node); - } - } - } - - if (posts_list && posts_list->size) { - for (size_t i = 0; i < posts_list->size; i++) { - // For each MultiTargetNode, we're going to have one additional - // anonymous local not represented in the locals table. We want - // to account for this in our table size. - const pm_node_t *post = posts_list->nodes[i]; - - if (PM_NODE_TYPE_P(post, PM_MULTI_TARGET_NODE)) { - PUSH_GETLOCAL(ret, location, table_size - body->param.post_start - (int) i, 0); - pm_compile_destructured_param_writes(iseq, (const pm_multi_target_node_t *) post, ret, scope_node); - } - } - } - - switch (body->type) { - case ISEQ_TYPE_BLOCK: { - LABEL *start = ISEQ_COMPILE_DATA(iseq)->start_label = NEW_LABEL(0); - LABEL *end = ISEQ_COMPILE_DATA(iseq)->end_label = NEW_LABEL(0); - const pm_node_location_t block_location = { .line = body->location.first_lineno, .node_id = -1 }; - - start->rescued = LABEL_RESCUE_BEG; - end->rescued = LABEL_RESCUE_END; - - // For nodes automatically assign the iteration variable to whatever - // index variable. We need to handle that write here because it has - // to happen in the context of the block. Note that this happens - // before the B_CALL tracepoint event. - if (PM_NODE_TYPE_P(scope_node->ast_node, PM_FOR_NODE)) { - pm_compile_for_node_index(iseq, ((const pm_for_node_t *) scope_node->ast_node)->index, ret, scope_node); - } - - PUSH_TRACE(ret, RUBY_EVENT_B_CALL); - PUSH_INSN(ret, block_location, nop); - PUSH_LABEL(ret, start); - - if (scope_node->body != NULL) { - switch (PM_NODE_TYPE(scope_node->ast_node)) { - case PM_POST_EXECUTION_NODE: { - const pm_post_execution_node_t *cast = (const pm_post_execution_node_t *) scope_node->ast_node; - PUSH_INSN1(ret, block_location, putspecialobject, INT2FIX(VM_SPECIAL_OBJECT_VMCORE)); - - // We create another ScopeNode from the statements within the PostExecutionNode - pm_scope_node_t next_scope_node; - pm_scope_node_init((const pm_node_t *) cast->statements, &next_scope_node, scope_node); - - const rb_iseq_t *block = NEW_CHILD_ISEQ(&next_scope_node, make_name_for_block(body->parent_iseq), ISEQ_TYPE_BLOCK, location.line); - pm_scope_node_destroy(&next_scope_node); - - PUSH_CALL_WITH_BLOCK(ret, block_location, id_core_set_postexe, INT2FIX(0), block); - break; - } - case PM_INTERPOLATED_REGULAR_EXPRESSION_NODE: { - const pm_interpolated_regular_expression_node_t *cast = (const pm_interpolated_regular_expression_node_t *) scope_node->ast_node; - pm_compile_regexp_dynamic(iseq, (const pm_node_t *) cast, &cast->parts, &location, ret, popped, scope_node); - break; - } - default: - pm_compile_node(iseq, scope_node->body, ret, popped, scope_node); - break; - } - } - else { - PUSH_INSN(ret, block_location, putnil); - } - - PUSH_LABEL(ret, end); - PUSH_TRACE(ret, RUBY_EVENT_B_RETURN); - ISEQ_COMPILE_DATA(iseq)->last_line = body->location.code_location.end_pos.lineno; - - /* wide range catch handler must put at last */ - PUSH_CATCH_ENTRY(CATCH_TYPE_REDO, start, end, NULL, start); - PUSH_CATCH_ENTRY(CATCH_TYPE_NEXT, start, end, NULL, end); - break; - } - case ISEQ_TYPE_ENSURE: { - const pm_node_location_t statements_location = (scope_node->body != NULL ? PM_NODE_START_LOCATION(scope_node->parser, scope_node->body) : location); - iseq_set_exception_local_table(iseq); - - if (scope_node->body != NULL) { - PM_COMPILE_POPPED((const pm_node_t *) scope_node->body); - } - - PUSH_GETLOCAL(ret, statements_location, 1, 0); - PUSH_INSN1(ret, statements_location, throw, INT2FIX(0)); - return; - } - case ISEQ_TYPE_METHOD: { - ISEQ_COMPILE_DATA(iseq)->root_node = (const void *) scope_node->body; - PUSH_TRACE(ret, RUBY_EVENT_CALL); - - if (scope_node->body) { - PM_COMPILE((const pm_node_t *) scope_node->body); - } - else { - PUSH_INSN(ret, location, putnil); - } - - ISEQ_COMPILE_DATA(iseq)->root_node = (const void *) scope_node->body; - PUSH_TRACE(ret, RUBY_EVENT_RETURN); - - ISEQ_COMPILE_DATA(iseq)->last_line = body->location.code_location.end_pos.lineno; - break; - } - case ISEQ_TYPE_RESCUE: { - iseq_set_exception_local_table(iseq); - if (PM_NODE_TYPE_P(scope_node->ast_node, PM_RESCUE_MODIFIER_NODE)) { - LABEL *lab = NEW_LABEL(lineno); - LABEL *rescue_end = NEW_LABEL(lineno); - PUSH_GETLOCAL(ret, location, LVAR_ERRINFO, 0); - PUSH_INSN1(ret, location, putobject, rb_eStandardError); - PUSH_INSN1(ret, location, checkmatch, INT2FIX(VM_CHECKMATCH_TYPE_RESCUE)); - PUSH_INSNL(ret, location, branchif, lab); - PUSH_INSNL(ret, location, jump, rescue_end); - PUSH_LABEL(ret, lab); - PUSH_TRACE(ret, RUBY_EVENT_RESCUE); - PM_COMPILE((const pm_node_t *) scope_node->body); - PUSH_INSN(ret, location, leave); - PUSH_LABEL(ret, rescue_end); - PUSH_GETLOCAL(ret, location, LVAR_ERRINFO, 0); - } - else { - PM_COMPILE((const pm_node_t *) scope_node->ast_node); - } - PUSH_INSN1(ret, location, throw, INT2FIX(0)); - - return; - } - default: - if (scope_node->body) { - PM_COMPILE((const pm_node_t *) scope_node->body); - } - else { - PUSH_INSN(ret, location, putnil); - } - break; - } - - if (PM_NODE_TYPE_P(scope_node->ast_node, PM_CLASS_NODE) || PM_NODE_TYPE_P(scope_node->ast_node, PM_MODULE_NODE)) { - const pm_node_location_t end_location = PM_NODE_END_LOCATION(scope_node->parser, scope_node->ast_node); - ADD_TRACE(ret, RUBY_EVENT_END); - ISEQ_COMPILE_DATA(iseq)->last_line = end_location.line; - } - - if (!PM_NODE_TYPE_P(scope_node->ast_node, PM_ENSURE_NODE)) { - const pm_node_location_t location = { .line = ISEQ_COMPILE_DATA(iseq)->last_line, .node_id = -1 }; - PUSH_INSN(ret, location, leave); - } - + case PM_SCOPE_NODE: + pm_compile_scope_node(iseq, (pm_scope_node_t *) node, &location, ret, popped); return; - } case PM_SELF_NODE: { // self // ^^^^ @@ -9403,7 +10334,7 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, pm_compile_constant_path_operator_write_node(iseq, (const pm_constant_path_operator_write_node_t *) cast->write, shareability, &location, ret, popped, scope_node); break; default: - rb_bug("Unexpected node type for shareable constant write: %s", pm_node_type_to_str(PM_NODE_TYPE(cast->write))); + rb_bug("Unexpected node type for shareable constant write: %s", pm_node_type(PM_NODE_TYPE(cast->write))); break; } @@ -9424,7 +10355,17 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, ID singletonclass; CONST_ID(singletonclass, "singletonclass"); - PUSH_INSN3(ret, location, defineclass, ID2SYM(singletonclass), child_iseq, INT2FIX(VM_DEFINECLASS_TYPE_SINGLETON_CLASS)); + + /* `class << self` in a class body and `class << Foo` (constant + receiver) are stable. All other forms are potentially dynamic. */ + int sclass_flags = VM_DEFINECLASS_TYPE_SINGLETON_CLASS; + if (!(PM_NODE_TYPE_P(cast->expression, PM_SELF_NODE) && + ISEQ_BODY(iseq)->type == ISEQ_TYPE_CLASS) && + !pm_cpath_const_p(cast->expression)) { + sclass_flags |= VM_DEFINECLASS_FLAG_DYNAMIC_CREF; + } + + PUSH_INSN3(ret, location, defineclass, ID2SYM(singletonclass), child_iseq, INT2FIX(sclass_flags)); if (popped) PUSH_INSN(ret, location, pop); RB_OBJ_WRITTEN(iseq, Qundef, (VALUE) child_iseq); @@ -9451,10 +10392,10 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, PUSH_INSN1(ret, location, putobject, string); } else if (PM_NODE_FLAG_P(cast, PM_STRING_FLAGS_MUTABLE)) { - PUSH_INSN1(ret, location, putstring, string); + PUSH_INSN1(ret, location, dupstring, string); } else { - PUSH_INSN1(ret, location, putchilledstring, string); + PUSH_INSN1(ret, location, dupchilledstring, string); } } return; @@ -9508,65 +10449,20 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, PUSH_INSN1(ret, location, putobject, value); } else if (PM_NODE_FLAG_P(node, PM_STRING_FLAGS_MUTABLE)) { - PUSH_INSN1(ret, location, putstring, value); + PUSH_INSN1(ret, location, dupstring, value); } else { - PUSH_INSN1(ret, location, putchilledstring, value); + PUSH_INSN1(ret, location, dupchilledstring, value); } } return; } - case PM_SUPER_NODE: { + case PM_SUPER_NODE: + // super() // super(foo) - // ^^^^^^^^^^ - const pm_super_node_t *cast = (const pm_super_node_t *) node; - - DECL_ANCHOR(args); - INIT_ANCHOR(args); - - LABEL *retry_label = NEW_LABEL(location.line); - LABEL *retry_end_l = NEW_LABEL(location.line); - - const rb_iseq_t *previous_block = ISEQ_COMPILE_DATA(iseq)->current_block; - const rb_iseq_t *current_block; - ISEQ_COMPILE_DATA(iseq)->current_block = current_block = NULL; - - PUSH_LABEL(ret, retry_label); - PUSH_INSN(ret, location, putself); - - int flags = 0; - struct rb_callinfo_kwarg *keywords = NULL; - int argc = pm_setup_args(cast->arguments, cast->block, &flags, &keywords, iseq, ret, scope_node, &location); - flags |= VM_CALL_SUPER | VM_CALL_FCALL; - - if (cast->block && PM_NODE_TYPE_P(cast->block, PM_BLOCK_NODE)) { - pm_scope_node_t next_scope_node; - pm_scope_node_init(cast->block, &next_scope_node, scope_node); - - ISEQ_COMPILE_DATA(iseq)->current_block = current_block = NEW_CHILD_ISEQ(&next_scope_node, make_name_for_block(iseq), ISEQ_TYPE_BLOCK, lineno); - pm_scope_node_destroy(&next_scope_node); - } - - if ((flags & VM_CALL_ARGS_BLOCKARG) && (flags & VM_CALL_KW_SPLAT) && !(flags & VM_CALL_KW_SPLAT_MUT)) { - PUSH_INSN(args, location, splatkw); - } - - PUSH_SEQ(ret, args); - if (ISEQ_BODY(ISEQ_BODY(iseq)->local_iseq)->param.flags.forwardable) { - flags |= VM_CALL_FORWARDING; - PUSH_INSN2(ret, location, invokesuperforward, new_callinfo(iseq, 0, argc, flags, keywords, current_block != NULL), current_block); - } - else { - PUSH_INSN2(ret, location, invokesuper, new_callinfo(iseq, 0, argc, flags, keywords, current_block != NULL), current_block); - pm_compile_retry_end_label(iseq, ret, retry_end_l); - } - - if (popped) PUSH_INSN(ret, location, pop); - ISEQ_COMPILE_DATA(iseq)->current_block = previous_block; - PUSH_CATCH_ENTRY(CATCH_TYPE_BREAK, retry_label, retry_end_l, current_block, retry_end_l); - + // super(...) + pm_compile_super_node(iseq, (const pm_super_node_t *) node, &location, ret, popped, scope_node); return; - } case PM_SYMBOL_NODE: { // :foo // ^^^^ @@ -9612,12 +10508,12 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, // bar unless foo // ^^^^^^^^^^^^^^ const pm_unless_node_t *cast = (const pm_unless_node_t *) node; - const pm_statements_node_t *consequent = NULL; - if (cast->consequent != NULL) { - consequent = ((const pm_else_node_t *) cast->consequent)->statements; + const pm_statements_node_t *statements = NULL; + if (cast->else_clause != NULL) { + statements = ((const pm_else_node_t *) cast->else_clause)->statements; } - pm_compile_conditional(iseq, &location, PM_UNLESS_NODE, (const pm_node_t *) cast, consequent, (const pm_node_t *) cast->statements, cast->predicate, ret, popped, scope_node); + pm_compile_conditional(iseq, &location, PM_UNLESS_NODE, (const pm_node_t *) cast, statements, (const pm_node_t *) cast->statements, cast->predicate, ret, popped, scope_node); return; } case PM_UNTIL_NODE: { @@ -9653,49 +10549,22 @@ pm_compile_node(rb_iseq_t *iseq, const pm_node_t *node, LINK_ANCHOR *const ret, return; } - case PM_YIELD_NODE: { + case PM_YIELD_NODE: // yield // ^^^^^ // // yield 1 // ^^^^^^^ - const pm_yield_node_t *cast = (const pm_yield_node_t *) node; - - switch (ISEQ_BODY(ISEQ_BODY(iseq)->local_iseq)->type) { - case ISEQ_TYPE_TOP: - case ISEQ_TYPE_MAIN: - case ISEQ_TYPE_CLASS: - COMPILE_ERROR(iseq, location.line, "Invalid yield"); - return; - default: /* valid */; - } - - int argc = 0; - int flags = 0; - struct rb_callinfo_kwarg *keywords = NULL; - - if (cast->arguments) { - argc = pm_setup_args(cast->arguments, NULL, &flags, &keywords, iseq, ret, scope_node, &location); - } - - PUSH_INSN1(ret, location, invokeblock, new_callinfo(iseq, 0, argc, flags, keywords, FALSE)); - if (popped) PUSH_INSN(ret, location, pop); - - int level = 0; - for (const rb_iseq_t *tmp_iseq = iseq; tmp_iseq != ISEQ_BODY(iseq)->local_iseq; level++) { - tmp_iseq = ISEQ_BODY(tmp_iseq)->parent_iseq; - } - - if (level > 0) access_outer_variables(iseq, level, rb_intern("yield"), true); + pm_compile_yield_node(iseq, (const pm_yield_node_t *) node, &location, ret, popped, scope_node); return; - } - default: { - rb_raise(rb_eNotImpError, "node type %s not implemented", pm_node_type_to_str(PM_NODE_TYPE(node))); + default: + rb_raise(rb_eNotImpError, "node type %s not implemented", pm_node_type(PM_NODE_TYPE(node))); return; - } } } +#undef PM_CONTAINER_P + /** True if the given iseq can have pre execution blocks. */ static inline bool pm_iseq_pre_execution_p(rb_iseq_t *iseq) @@ -9721,7 +10590,6 @@ VALUE pm_iseq_compile_node(rb_iseq_t *iseq, pm_scope_node_t *node) { DECL_ANCHOR(ret); - INIT_ANCHOR(ret); if (pm_iseq_pre_execution_p(iseq)) { // Because these ISEQs can have BEGIN{}, we're going to create two @@ -9729,14 +10597,12 @@ pm_iseq_compile_node(rb_iseq_t *iseq, pm_scope_node_t *node) // on the scope node so that when BEGIN{} is found, its contents will be // added to the "pre" anchor. DECL_ANCHOR(pre); - INIT_ANCHOR(pre); node->pre_execution_anchor = pre; // Now we'll compile the body as normal. We won't compile directly into // the "ret" anchor yet because we want to add the "pre" anchor to the // beginning of the "ret" anchor first. DECL_ANCHOR(body); - INIT_ANCHOR(body); pm_compile_node(iseq, (const pm_node_t *) node, body, false, node); // Now we'll join both anchors together so that the content is in the @@ -9754,6 +10620,15 @@ pm_iseq_compile_node(rb_iseq_t *iseq, pm_scope_node_t *node) return iseq_setup(iseq, ret); } +void +pm_parse_result_init(pm_parse_result_t *result) +{ + memset(result, 0, sizeof(pm_parse_result_t)); + result->arena = pm_arena_new(); + result->options = pm_options_new(); + pm_options_line_set(result->options, 1); +} + /** * Free the internal memory associated with a pm_parse_result_t struct. * Importantly this does not free the struct itself. @@ -9762,21 +10637,20 @@ void pm_parse_result_free(pm_parse_result_t *result) { if (result->parsed) { - pm_node_destroy(&result->parser, result->node.ast_node); - - xfree(result->node.constants); + SIZED_FREE_N(result->node.constants, pm_parser_constants_size(result->node.parser)); pm_scope_node_destroy(&result->node); } - pm_parser_free(&result->parser); - pm_string_free(&result->input); - pm_options_free(&result->options); + if (result->parser) pm_parser_free(result->parser); + pm_arena_free(result->arena); + if (result->source) pm_source_free(result->source); + pm_options_free(result->options); } /** An error that is going to be formatted into the output. */ typedef struct { /** A pointer to the diagnostic that was generated during parsing. */ - pm_diagnostic_t *error; + const pm_diagnostic_t *error; /** The start line of the diagnostic message. */ int32_t line; @@ -9806,125 +10680,157 @@ typedef struct { size_t divider_length; } pm_parse_error_format_t; -#define PM_COLOR_GRAY "\033[38;5;102m" +#define PM_COLOR_BOLD "\033[1m" +#define PM_COLOR_GRAY "\033[2m" #define PM_COLOR_RED "\033[1;31m" #define PM_COLOR_RESET "\033[m" #define PM_ERROR_TRUNCATE 30 -static inline pm_parse_error_t * -pm_parse_errors_format_sort(const pm_parser_t *parser, const pm_list_t *error_list, const pm_newline_list_t *newline_list) { - pm_parse_error_t *errors = xcalloc(error_list->size, sizeof(pm_parse_error_t)); - if (errors == NULL) return NULL; +/** Context struct for collecting errors via callback. */ +typedef struct { + pm_parse_error_t *errors; + size_t count; + size_t capacity; + const pm_line_offset_list_t *line_offsets; + int32_t start_line; +} pm_error_collect_t; - int32_t start_line = parser->start_line; - for (pm_diagnostic_t *error = (pm_diagnostic_t *) error_list->head; error != NULL; error = (pm_diagnostic_t *) error->node.next) { - pm_line_column_t start = pm_newline_list_line_column(newline_list, error->location.start, start_line); - pm_line_column_t end = pm_newline_list_line_column(newline_list, error->location.end, start_line); - - // We're going to insert this error into the array in sorted order. We - // do this by finding the first error that has a line number greater - // than the current error and then inserting the current error before - // that one. - size_t index = 0; - while ( - (index < error_list->size) && - (errors[index].error != NULL) && - ( - (errors[index].line < start.line) || - ((errors[index].line == start.line) && (errors[index].column_start < start.column)) - ) - ) index++; - - // Now we're going to shift all of the errors after this one down one - // index to make room for the new error. - if (index + 1 < error_list->size) { - memmove(&errors[index + 1], &errors[index], sizeof(pm_parse_error_t) * (error_list->size - index - 1)); - } - - // Finally, we'll insert the error into the array. - uint32_t column_end; - if (start.line == end.line) { - column_end = end.column; - } else { - column_end = (uint32_t) (newline_list->offsets[start.line - start_line + 1] - newline_list->offsets[start.line - start_line] - 1); - } +static void +pm_error_collect_callback(const pm_diagnostic_t *diagnostic, void *data) +{ + pm_error_collect_t *ctx = (pm_error_collect_t *) data; + pm_location_t loc = pm_diagnostic_location(diagnostic); - // Ensure we have at least one column of error. - if (start.column == column_end) column_end++; + pm_line_column_t start = pm_line_offset_list_line_column(ctx->line_offsets, loc.start, ctx->start_line); + pm_line_column_t end = pm_line_offset_list_line_column(ctx->line_offsets, loc.start + loc.length, ctx->start_line); - errors[index] = (pm_parse_error_t) { - .error = error, - .line = start.line, - .column_start = start.column, - .column_end = column_end - }; + uint32_t column_end; + if (start.line == end.line) { + column_end = end.column; + } else { + column_end = (uint32_t) (ctx->line_offsets->offsets[start.line - ctx->start_line + 1] - ctx->line_offsets->offsets[start.line - ctx->start_line] - 1); + } + + // Ensure we have at least one column of error. + if (start.column == column_end) column_end++; + + // Insert into sorted position (insertion sort). + size_t index = 0; + while ( + (index < ctx->count) && + ( + (ctx->errors[index].line < start.line) || + ((ctx->errors[index].line == start.line) && (ctx->errors[index].column_start < start.column)) + ) + ) index++; + + if (index < ctx->count) { + memmove(&ctx->errors[index + 1], &ctx->errors[index], sizeof(pm_parse_error_t) * (ctx->count - index)); } + ctx->errors[index] = (pm_parse_error_t) { + .error = diagnostic, + .line = start.line, + .column_start = start.column, + .column_end = column_end + }; + ctx->count++; +} + +static inline pm_parse_error_t * +pm_parse_errors_format_sort(const pm_parser_t *parser, size_t error_count, const pm_line_offset_list_t *line_offsets) { + pm_parse_error_t *errors = xcalloc(error_count, sizeof(pm_parse_error_t)); + if (errors == NULL) return NULL; + + pm_error_collect_t ctx = { + .errors = errors, + .count = 0, + .capacity = error_count, + .line_offsets = line_offsets, + .start_line = pm_parser_start_line(parser) + }; + + pm_parser_errors_each(parser, pm_error_collect_callback, &ctx); + return errors; } static inline void -pm_parse_errors_format_line(const pm_parser_t *parser, const pm_newline_list_t *newline_list, const char *number_prefix, int32_t line, uint32_t column_start, uint32_t column_end, pm_buffer_t *buffer) { - int32_t line_delta = line - parser->start_line; +pm_parse_errors_format_line(const pm_parser_t *parser, const pm_line_offset_list_t *line_offsets, const char *number_prefix, int32_t line, uint32_t column_start, uint32_t column_end, VALUE buffer) { + int32_t line_delta = line - pm_parser_start_line(parser); assert(line_delta >= 0); size_t index = (size_t) line_delta; - assert(index < newline_list->size); + assert(index < line_offsets->size); - const uint8_t *start = &parser->start[newline_list->offsets[index]]; + const uint8_t *start = &pm_parser_start(parser)[line_offsets->offsets[index]]; const uint8_t *end; - if (index >= newline_list->size - 1) { - end = parser->end; + if (index >= line_offsets->size - 1) { + end = pm_parser_end(parser); } else { - end = &parser->start[newline_list->offsets[index + 1]]; + end = &pm_parser_start(parser)[line_offsets->offsets[index + 1]]; } - pm_buffer_append_format(buffer, number_prefix, line); + rb_str_catf(buffer, number_prefix, line); // Here we determine if we should truncate the end of the line. bool truncate_end = false; if ((column_end != 0) && ((end - (start + column_end)) >= PM_ERROR_TRUNCATE)) { - end = start + column_end + PM_ERROR_TRUNCATE; + const uint8_t *end_candidate = start + column_end + PM_ERROR_TRUNCATE; + + for (const uint8_t *ptr = start; ptr < end_candidate;) { + size_t char_width = pm_parser_encoding_char_width(parser, ptr, pm_parser_end(parser) - ptr); + + // If we failed to decode a character, then just bail out and + // truncate at the fixed width. + if (char_width == 0) break; + + // If this next character would go past the end candidate, + // then we need to truncate before it. + if (ptr + char_width > end_candidate) { + end_candidate = ptr; + break; + } + + ptr += char_width; + } + + end = end_candidate; truncate_end = true; } // Here we determine if we should truncate the start of the line. if (column_start >= PM_ERROR_TRUNCATE) { - pm_buffer_append_string(buffer, "... ", 4); + rb_str_cat(buffer, "... ", 4); start += column_start; } - pm_buffer_append_string(buffer, (const char *) start, (size_t) (end - start)); + rb_str_cat(buffer, (const char *) start, (size_t) (end - start)); if (truncate_end) { - pm_buffer_append_string(buffer, " ...\n", 5); - } else if (end == parser->end && end[-1] != '\n') { - pm_buffer_append_string(buffer, "\n", 1); + rb_str_cat(buffer, " ...\n", 5); + } else if (end == pm_parser_end(parser) && end[-1] != '\n') { + rb_str_cat(buffer, "\n", 1); } } /** - * Format the errors on the parser into the given buffer. + * Format a pre-sorted array of errors into the given buffer. */ static void -pm_parse_errors_format(const pm_parser_t *parser, const pm_list_t *error_list, pm_buffer_t *buffer, bool colorize, bool inline_messages) { - assert(error_list->size != 0); - - // First, we're going to sort all of the errors by line number using an - // insertion sort into a newly allocated array. - const int32_t start_line = parser->start_line; - const pm_newline_list_t *newline_list = &parser->newline_list; +pm_parse_errors_format_with(const pm_parser_t *parser, pm_parse_error_t *errors, size_t error_count, VALUE buffer, int highlight, bool inline_messages) { + assert(error_count != 0); - pm_parse_error_t *errors = pm_parse_errors_format_sort(parser, error_list, newline_list); - if (errors == NULL) return; + const int32_t start_line = pm_parser_start_line(parser); + const pm_line_offset_list_t *line_offsets = pm_parser_line_offsets(parser); // Now we're going to determine how we're going to format line numbers and // blank lines based on the maximum number of digits in the line numbers // that are going to be displaid. pm_parse_error_format_t error_format; int32_t first_line_number = errors[0].line; - int32_t last_line_number = errors[error_list->size - 1].line; + int32_t last_line_number = errors[error_count - 1].line; // If we have a maximum line number that is negative, then we're going to // use the absolute value for comparison but multiple by 10 to additionally @@ -9934,7 +10840,7 @@ pm_parse_errors_format(const pm_parser_t *parser, const pm_list_t *error_list, p int32_t max_line_number = first_line_number > last_line_number ? first_line_number : last_line_number; if (max_line_number < 10) { - if (colorize) { + if (highlight > 0) { error_format = (pm_parse_error_format_t) { .number_prefix = PM_COLOR_GRAY "%1" PRIi32 " | " PM_COLOR_RESET, .blank_prefix = PM_COLOR_GRAY " | " PM_COLOR_RESET, @@ -9948,7 +10854,7 @@ pm_parse_errors_format(const pm_parser_t *parser, const pm_list_t *error_list, p }; } } else if (max_line_number < 100) { - if (colorize) { + if (highlight > 0) { error_format = (pm_parse_error_format_t) { .number_prefix = PM_COLOR_GRAY "%2" PRIi32 " | " PM_COLOR_RESET, .blank_prefix = PM_COLOR_GRAY " | " PM_COLOR_RESET, @@ -9962,7 +10868,7 @@ pm_parse_errors_format(const pm_parser_t *parser, const pm_list_t *error_list, p }; } } else if (max_line_number < 1000) { - if (colorize) { + if (highlight > 0) { error_format = (pm_parse_error_format_t) { .number_prefix = PM_COLOR_GRAY "%3" PRIi32 " | " PM_COLOR_RESET, .blank_prefix = PM_COLOR_GRAY " | " PM_COLOR_RESET, @@ -9976,7 +10882,7 @@ pm_parse_errors_format(const pm_parser_t *parser, const pm_list_t *error_list, p }; } } else if (max_line_number < 10000) { - if (colorize) { + if (highlight > 0) { error_format = (pm_parse_error_format_t) { .number_prefix = PM_COLOR_GRAY "%4" PRIi32 " | " PM_COLOR_RESET, .blank_prefix = PM_COLOR_GRAY " | " PM_COLOR_RESET, @@ -9990,7 +10896,7 @@ pm_parse_errors_format(const pm_parser_t *parser, const pm_list_t *error_list, p }; } } else { - if (colorize) { + if (highlight > 0) { error_format = (pm_parse_error_format_t) { .number_prefix = PM_COLOR_GRAY "%5" PRIi32 " | " PM_COLOR_RESET, .blank_prefix = PM_COLOR_GRAY " | " PM_COLOR_RESET, @@ -10013,11 +10919,10 @@ pm_parse_errors_format(const pm_parser_t *parser, const pm_list_t *error_list, p // the source before the error to give some context. We'll be careful not to // display the same line twice in case the errors are close enough in the // source. - int32_t last_line = parser->start_line - 1; + int32_t last_line = pm_parser_start_line(parser) - 1; uint32_t last_column_start = 0; - const pm_encoding_t *encoding = parser->encoding; - for (size_t index = 0; index < error_list->size; index++) { + for (size_t index = 0; index < error_count; index++) { pm_parse_error_t *error = &errors[index]; // Here we determine how many lines of padding of the source to display, @@ -10025,40 +10930,42 @@ pm_parse_errors_format(const pm_parser_t *parser, const pm_list_t *error_list, p if (error->line - last_line > 1) { if (error->line - last_line > 2) { if ((index != 0) && (error->line - last_line > 3)) { - pm_buffer_append_string(buffer, error_format.divider, error_format.divider_length); + rb_str_cat(buffer, error_format.divider, error_format.divider_length); } - pm_buffer_append_string(buffer, " ", 2); - pm_parse_errors_format_line(parser, newline_list, error_format.number_prefix, error->line - 2, 0, 0, buffer); + rb_str_cat(buffer, " ", 2); + pm_parse_errors_format_line(parser, line_offsets, error_format.number_prefix, error->line - 2, 0, 0, buffer); } - pm_buffer_append_string(buffer, " ", 2); - pm_parse_errors_format_line(parser, newline_list, error_format.number_prefix, error->line - 1, 0, 0, buffer); + rb_str_cat(buffer, " ", 2); + pm_parse_errors_format_line(parser, line_offsets, error_format.number_prefix, error->line - 1, 0, 0, buffer); } // If this is the first error or we're on a new line, then we'll display // the line that has the error in it. if ((index == 0) || (error->line != last_line)) { - if (colorize) { - pm_buffer_append_string(buffer, PM_COLOR_RED "> " PM_COLOR_RESET, 12); + if (highlight > 1) { + rb_str_cat_cstr(buffer, PM_COLOR_RED "> " PM_COLOR_RESET); + } else if (highlight > 0) { + rb_str_cat_cstr(buffer, PM_COLOR_BOLD "> " PM_COLOR_RESET); } else { - pm_buffer_append_string(buffer, "> ", 2); + rb_str_cat_cstr(buffer, "> "); } last_column_start = error->column_start; // Find the maximum column end of all the errors on this line. uint32_t column_end = error->column_end; - for (size_t next_index = index + 1; next_index < error_list->size; next_index++) { + for (size_t next_index = index + 1; next_index < error_count; next_index++) { if (errors[next_index].line != error->line) break; if (errors[next_index].column_end > column_end) column_end = errors[next_index].column_end; } - pm_parse_errors_format_line(parser, newline_list, error_format.number_prefix, error->line, error->column_start, column_end, buffer); + pm_parse_errors_format_line(parser, line_offsets, error_format.number_prefix, error->line, error->column_start, column_end, buffer); } - const uint8_t *start = &parser->start[newline_list->offsets[error->line - start_line]]; - if (start == parser->end) pm_buffer_append_byte(buffer, '\n'); + const uint8_t *start = &pm_parser_start(parser)[line_offsets->offsets[error->line - start_line]]; + if (start == pm_parser_end(parser)) rb_str_cat(buffer, "\n", 1); // Now we'll display the actual error message. We'll do this by first // putting the prefix to the line, then a bunch of blank spaces @@ -10069,65 +10976,91 @@ pm_parse_errors_format(const pm_parser_t *parser, const pm_list_t *error_list, p // character when displaid in the terminal. For some east-asian // languages or emoji, this means it can be thrown off pretty badly. We // will need to solve this eventually. - pm_buffer_append_string(buffer, " ", 2); - pm_buffer_append_string(buffer, error_format.blank_prefix, error_format.blank_prefix_length); + rb_str_cat(buffer, " ", 2); + rb_str_cat(buffer, error_format.blank_prefix, error_format.blank_prefix_length); size_t column = 0; if (last_column_start >= PM_ERROR_TRUNCATE) { - pm_buffer_append_string(buffer, " ", 4); + rb_str_cat(buffer, " ", 4); column = last_column_start; } while (column < error->column_start) { - pm_buffer_append_byte(buffer, ' '); + rb_str_cat(buffer, " ", 1); - size_t char_width = encoding->char_width(start + column, parser->end - (start + column)); + size_t char_width = pm_parser_encoding_char_width(parser, start + column, pm_parser_end(parser) - (start + column)); column += (char_width == 0 ? 1 : char_width); } - if (colorize) pm_buffer_append_string(buffer, PM_COLOR_RED, 7); - pm_buffer_append_byte(buffer, '^'); + if (highlight > 1) rb_str_cat_cstr(buffer, PM_COLOR_RED); + else if (highlight > 0) rb_str_cat_cstr(buffer, PM_COLOR_BOLD); + rb_str_cat(buffer, "^", 1); - size_t char_width = encoding->char_width(start + column, parser->end - (start + column)); + size_t char_width = pm_parser_encoding_char_width(parser, start + column, pm_parser_end(parser) - (start + column)); column += (char_width == 0 ? 1 : char_width); while (column < error->column_end) { - pm_buffer_append_byte(buffer, '~'); + rb_str_cat(buffer, "~", 1); - size_t char_width = encoding->char_width(start + column, parser->end - (start + column)); + size_t char_width = pm_parser_encoding_char_width(parser, start + column, pm_parser_end(parser) - (start + column)); column += (char_width == 0 ? 1 : char_width); } - if (colorize) pm_buffer_append_string(buffer, PM_COLOR_RESET, 3); + if (highlight > 0) rb_str_cat_cstr(buffer, PM_COLOR_RESET); if (inline_messages) { - pm_buffer_append_byte(buffer, ' '); + rb_str_cat(buffer, " ", 1); assert(error->error != NULL); - const char *message = error->error->message; - pm_buffer_append_string(buffer, message, strlen(message)); + const char *message = pm_diagnostic_message(error->error); + rb_str_cat(buffer, message, strlen(message)); } - pm_buffer_append_byte(buffer, '\n'); + rb_str_cat(buffer, "\n", 1); // Here we determine how many lines of padding to display after the // error, depending on where the next error is in source. last_line = error->line; - int32_t next_line = (index == error_list->size - 1) ? (((int32_t) newline_list->size) + parser->start_line) : errors[index + 1].line; + int32_t next_line; + + if (index == error_count - 1) { + next_line = (((int32_t) line_offsets->size) + pm_parser_start_line(parser)); + + // If the file ends with a newline, subtract one from our "next_line" + // so that we don't output an extra line at the end of the file + if ((pm_parser_start(parser) + line_offsets->offsets[line_offsets->size - 1]) == pm_parser_end(parser)) { + next_line--; + } + } + else { + next_line = errors[index + 1].line; + } if (next_line - last_line > 1) { - pm_buffer_append_string(buffer, " ", 2); - pm_parse_errors_format_line(parser, newline_list, error_format.number_prefix, ++last_line, 0, 0, buffer); + rb_str_cat(buffer, " ", 2); + pm_parse_errors_format_line(parser, line_offsets, error_format.number_prefix, ++last_line, 0, 0, buffer); } if (next_line - last_line > 1) { - pm_buffer_append_string(buffer, " ", 2); - pm_parse_errors_format_line(parser, newline_list, error_format.number_prefix, ++last_line, 0, 0, buffer); + rb_str_cat(buffer, " ", 2); + pm_parse_errors_format_line(parser, line_offsets, error_format.number_prefix, ++last_line, 0, 0, buffer); } } - // Finally, we'll free the array of errors that we allocated. - xfree(errors); +} + +/** + * Format the errors on the parser into the given buffer. + */ +static void +pm_parse_errors_format(const pm_parser_t *parser, size_t error_count, VALUE buffer, int highlight, bool inline_messages) { + const pm_line_offset_list_t *line_offsets = pm_parser_line_offsets(parser); + + pm_parse_error_t *errors = pm_parse_errors_format_sort(parser, error_count, line_offsets); + if (errors == NULL) return; + + pm_parse_errors_format_with(parser, errors, error_count, buffer, highlight, inline_messages); + SIZED_FREE_N(errors, error_count); } #undef PM_ERROR_TRUNCATE @@ -10142,130 +11075,256 @@ pm_parse_errors_format(const pm_parser_t *parser, const pm_list_t *error_list, p * as well. */ static bool -pm_parse_process_error_utf8_p(const pm_parser_t *parser, const pm_location_t *location) +pm_parse_process_error_utf8_p(const pm_parser_t *parser, pm_location_t location) { - const size_t start_line = pm_newline_list_line_column(&parser->newline_list, location->start, 1).line; - const size_t end_line = pm_newline_list_line_column(&parser->newline_list, location->end, 1).line; + const size_t start_line = pm_line_offset_list_line_column(pm_parser_line_offsets(parser), location.start, 1).line; + const size_t end_line = pm_line_offset_list_line_column(pm_parser_line_offsets(parser), location.start + location.length, 1).line; - const uint8_t *start = parser->start + parser->newline_list.offsets[start_line - 1]; - const uint8_t *end = ((end_line == parser->newline_list.size) ? parser->end : (parser->start + parser->newline_list.offsets[end_line])); - size_t width; + const pm_line_offset_list_t *line_offsets = pm_parser_line_offsets(parser); + const uint8_t *start = pm_parser_start(parser) + line_offsets->offsets[start_line - 1]; + const uint8_t *end = ((end_line == line_offsets->size) ? pm_parser_end(parser) : (pm_parser_start(parser) + line_offsets->offsets[end_line])); + rb_encoding *utf8 = rb_utf8_encoding(); while (start < end) { - if ((width = pm_encoding_utf_8_char_width(start, end - start)) == 0) return false; - start += width; + int width = rb_enc_precise_mbclen((const char *) start, (const char *) end, utf8); + if (!MBCLEN_CHARFOUND_P(width)) return false; + start += MBCLEN_CHARFOUND_LEN(width); } return true; } -/** - * Generate an error object from the given parser that contains as much - * information as possible about the errors that were encountered. - */ -static VALUE -pm_parse_process_error(const pm_parse_result_t *result) +/** Context for the error processing callback used in pm_parse_process_error. */ +typedef struct { + const pm_parse_result_t *result; + const pm_parser_t *parser; + const pm_string_t *filepath; + VALUE buffer; + int highlight; + bool valid_utf8; + bool found_argument_error; + bool found_load_error; + VALUE early_return; + const pm_diagnostic_t *first_error; + size_t error_count; +} pm_process_error_ctx_t; + +static void +pm_process_error_check_callback(const pm_diagnostic_t *diagnostic, void *data) { - const pm_parser_t *parser = &result->parser; - const pm_diagnostic_t *head = (const pm_diagnostic_t *) parser->error_list.head; - bool valid_utf8 = true; + pm_process_error_ctx_t *ctx = (pm_process_error_ctx_t *) data; + pm_location_t loc = pm_diagnostic_location(diagnostic); - pm_buffer_t buffer = { 0 }; - const pm_string_t *filepath = &parser->filepath; + if (ctx->first_error == NULL) ctx->first_error = diagnostic; + ctx->error_count++; - for (const pm_diagnostic_t *error = head; error != NULL; error = (const pm_diagnostic_t *) error->node.next) { - switch (error->level) { - case PM_ERROR_LEVEL_SYNTAX: - // It is implicitly assumed that the error messages will be - // encodeable as UTF-8. Because of this, we can't include source - // examples that contain invalid byte sequences. So if any source - // examples include invalid UTF-8 byte sequences, we will skip - // showing source examples entirely. - if (valid_utf8 && !pm_parse_process_error_utf8_p(parser, &error->location)) { - valid_utf8 = false; + switch (pm_diagnostic_error_level(diagnostic)) { + case PM_ERROR_LEVEL_SYNTAX: + if (ctx->valid_utf8 && !pm_parse_process_error_utf8_p(ctx->parser, loc)) { + ctx->valid_utf8 = false; + } + break; + case PM_ERROR_LEVEL_ARGUMENT: { + if (ctx->found_argument_error || ctx->found_load_error) break; + ctx->found_argument_error = true; + + int32_t line_number = (int32_t) pm_location_line_number(ctx->parser, &loc); + + rb_str_catf( + ctx->buffer, + "%.*s:%" PRIi32 ": %s", + (int) pm_string_length(ctx->filepath), + pm_string_source(ctx->filepath), + line_number, + pm_diagnostic_message(diagnostic) + ); + + if (pm_parse_process_error_utf8_p(ctx->parser, loc)) { + rb_str_cat(ctx->buffer, "\n", 1); + // Format just this one error. We construct a single-element sorted + // array manually and call the format function with count=1. + const pm_line_offset_list_t *line_offsets = pm_parser_line_offsets(ctx->parser); + int32_t start_line = pm_parser_start_line(ctx->parser); + pm_line_column_t start_lc = pm_line_offset_list_line_column(line_offsets, loc.start, start_line); + pm_line_column_t end_lc = pm_line_offset_list_line_column(line_offsets, loc.start + loc.length, start_line); + + uint32_t col_end; + if (start_lc.line == end_lc.line) { + col_end = end_lc.column; + } else { + col_end = (uint32_t) (line_offsets->offsets[start_lc.line - start_line + 1] - line_offsets->offsets[start_lc.line - start_line] - 1); } - break; - case PM_ERROR_LEVEL_ARGUMENT: { - // Any errors with the level PM_ERROR_LEVEL_ARGUMENT take over as - // the only argument that gets raised. This is to allow priority - // messages that should be handled before anything else. - int32_t line_number = (int32_t) pm_location_line_number(parser, &error->location); - - pm_buffer_append_format( - &buffer, - "%.*s:%" PRIi32 ": %s", - (int) pm_string_length(filepath), - pm_string_source(filepath), - line_number, - error->message - ); + if (start_lc.column == col_end) col_end++; - if (pm_parse_process_error_utf8_p(parser, &error->location)) { - pm_buffer_append_byte(&buffer, '\n'); + pm_parse_error_t single_error = { + .error = diagnostic, + .line = start_lc.line, + .column_start = start_lc.column, + .column_end = col_end + }; + pm_parse_errors_format_with(ctx->parser, &single_error, 1, ctx->buffer, ctx->highlight, false); + } - pm_list_node_t *list_node = (pm_list_node_t *) error; - pm_list_t error_list = { .size = 1, .head = list_node, .tail = list_node }; + ctx->early_return = rb_exc_new_str(rb_eArgError, ctx->buffer); + break; + } + case PM_ERROR_LEVEL_LOAD: { + if (ctx->found_argument_error || ctx->found_load_error) break; + ctx->found_load_error = true; - pm_parse_errors_format(parser, &error_list, &buffer, rb_stderr_tty_p(), false); - } + VALUE message = rb_enc_str_new_cstr(pm_diagnostic_message(diagnostic), rb_locale_encoding()); + VALUE value = rb_exc_new3(rb_eLoadError, message); + rb_ivar_set(value, rb_intern_const("@path"), Qnil); + ctx->early_return = value; + break; + } + } +} - VALUE value = rb_exc_new(rb_eArgError, pm_buffer_value(&buffer), pm_buffer_length(&buffer)); - pm_buffer_free(&buffer); +/** Callback for formatting non-UTF8 errors. */ +typedef struct { + const pm_parser_t *parser; + const pm_string_t *filepath; + VALUE buffer; + bool first; +} pm_error_simple_format_ctx_t; - return value; - } - case PM_ERROR_LEVEL_LOAD: { - // Load errors are much simpler, because they don't include any of - // the source in them. We create the error directly from the - // message. - VALUE message = rb_enc_str_new_cstr(error->message, rb_locale_encoding()); - VALUE value = rb_exc_new3(rb_eLoadError, message); - rb_ivar_set(value, rb_intern_const("@path"), Qnil); - return value; - } - } +static void +pm_error_simple_format_callback(const pm_diagnostic_t *diagnostic, void *data) +{ + pm_error_simple_format_ctx_t *ctx = (pm_error_simple_format_ctx_t *) data; + pm_location_t loc = pm_diagnostic_location(diagnostic); + + if (!ctx->first) rb_str_cat(ctx->buffer, "\n", 1); + ctx->first = false; + + rb_str_catf(ctx->buffer, "%.*s:%" PRIi32 ": %s", + (int) pm_string_length(ctx->filepath), + pm_string_source(ctx->filepath), + (int32_t) pm_location_line_number(ctx->parser, &loc), + pm_diagnostic_message(diagnostic)); +} + +/** + * Generate an error object from the given parser that contains as much + * information as possible about the errors that were encountered. + */ +static VALUE +pm_parse_process_error(const pm_parse_result_t *result) +{ + const pm_parser_t *parser = result->parser; + size_t error_count = pm_parser_errors_size(parser); + + VALUE buffer = rb_str_buf_new(0); + const pm_string_t *filepath = pm_parser_filepath(parser); + + int highlight = rb_stderr_tty_p(); + if (highlight) { + const char *no_color = getenv("NO_COLOR"); + highlight = (no_color == NULL || no_color[0] == '\0') ? 2 : 1; + } + + // First pass: check for argument/load errors and UTF-8 validity. + pm_process_error_ctx_t ctx = { + .result = result, + .parser = parser, + .filepath = filepath, + .buffer = buffer, + .highlight = highlight, + .valid_utf8 = true, + .found_argument_error = false, + .found_load_error = false, + .early_return = Qundef, + .first_error = NULL, + .error_count = 0 + }; + + pm_parser_errors_each(parser, pm_process_error_check_callback, &ctx); + + // If we found an argument or load error, return it immediately. + if (ctx.early_return != Qundef) { + return ctx.early_return; } - pm_buffer_append_format( - &buffer, + // Format the header line. + pm_location_t first_loc = pm_diagnostic_location(ctx.first_error); + rb_str_catf( + buffer, "%.*s:%" PRIi32 ": syntax error%s found\n", (int) pm_string_length(filepath), pm_string_source(filepath), - (int32_t) pm_location_line_number(parser, &head->location), - (parser->error_list.size > 1) ? "s" : "" + (int32_t) pm_location_line_number(parser, &first_loc), + (error_count > 1) ? "s" : "" ); - if (valid_utf8) { - pm_parse_errors_format(parser, &parser->error_list, &buffer, rb_stderr_tty_p(), true); + if (ctx.valid_utf8) { + pm_parse_errors_format(parser, error_count, buffer, highlight, true); } else { - for (const pm_diagnostic_t *error = head; error != NULL; error = (const pm_diagnostic_t *) error->node.next) { - if (error != head) pm_buffer_append_byte(&buffer, '\n'); - pm_buffer_append_format(&buffer, "%.*s:%" PRIi32 ": %s", (int) pm_string_length(filepath), pm_string_source(filepath), (int32_t) pm_location_line_number(parser, &error->location), error->message); - } + pm_error_simple_format_ctx_t simple_ctx = { + .parser = parser, + .filepath = filepath, + .buffer = buffer, + .first = true + }; + pm_parser_errors_each(parser, pm_error_simple_format_callback, &simple_ctx); } - VALUE message = rb_enc_str_new(pm_buffer_value(&buffer), pm_buffer_length(&buffer), result->node.encoding); - VALUE error = rb_exc_new_str(rb_eSyntaxError, message); + rb_enc_associate(buffer, result->node.encoding); + VALUE error = rb_exc_new_str(rb_eSyntaxError, buffer); rb_encoding *filepath_encoding = result->node.filepath_encoding != NULL ? result->node.filepath_encoding : rb_utf8_encoding(); VALUE path = rb_enc_str_new((const char *) pm_string_source(filepath), pm_string_length(filepath), filepath_encoding); rb_ivar_set(error, rb_intern_const("@path"), path); - pm_buffer_free(&buffer); return error; } +/** Context for interning constants via callback. */ +typedef struct { + ID *constants; + rb_encoding *encoding; + size_t index; +} pm_intern_constants_ctx_t; + +static void +pm_intern_constants_callback(const pm_constant_t *constant, void *data) +{ + pm_intern_constants_ctx_t *ctx = (pm_intern_constants_ctx_t *) data; + ctx->constants[ctx->index++] = rb_intern3((const char *) pm_constant_start(constant), pm_constant_length(constant), ctx->encoding); +} + +/** Context for emitting warnings via callback. */ +typedef struct { + const pm_parser_t *parser; + rb_encoding *encoding; + const char *filepath; +} pm_warning_emit_ctx_t; + +static void +pm_warning_emit_callback(const pm_diagnostic_t *diagnostic, void *data) { + pm_warning_emit_ctx_t *ctx = (pm_warning_emit_ctx_t *) data; + pm_location_t loc = pm_diagnostic_location(diagnostic); + int line = pm_location_line_number(ctx->parser, &loc); + + if (pm_diagnostic_warning_level(diagnostic) == PM_WARNING_LEVEL_VERBOSE) { + rb_enc_compile_warning(ctx->encoding, ctx->filepath, line, "%s", pm_diagnostic_message(diagnostic)); + } + else { + rb_enc_compile_warn(ctx->encoding, ctx->filepath, line, "%s", pm_diagnostic_message(diagnostic)); + } +} + /** * Parse the parse result and raise a Ruby error if there are any syntax errors. * It returns an error if one should be raised. It is assumed that the parse * result object is zeroed out. */ static VALUE -pm_parse_process(pm_parse_result_t *result, pm_node_t *node) +pm_parse_process(pm_parse_result_t *result, pm_node_t *node, VALUE *script_lines) { - pm_parser_t *parser = &result->parser; + pm_parser_t *parser = result->parser; // First, set up the scope node so that the AST node is attached and can be // freed regardless of whether or we return an error. @@ -10276,28 +11335,37 @@ pm_parse_process(pm_parse_result_t *result, pm_node_t *node) pm_scope_node_init(node, scope_node, NULL); scope_node->filepath_encoding = filepath_encoding; - scope_node->encoding = rb_enc_find(parser->encoding->name); - if (!scope_node->encoding) rb_bug("Encoding not found %s!", parser->encoding->name); + const char *encoding_name = pm_parser_encoding_name(parser); + scope_node->encoding = rb_enc_find(encoding_name); + if (!scope_node->encoding) rb_bug("Encoding not found %s!", encoding_name); scope_node->coverage_enabled = coverage_enabled; - // Emit all of the various warnings from the parse. - const pm_diagnostic_t *warning; - const char *warning_filepath = (const char *) pm_string_source(&parser->filepath); - - for (warning = (const pm_diagnostic_t *) parser->warning_list.head; warning != NULL; warning = (const pm_diagnostic_t *) warning->node.next) { - int line = pm_location_line_number(parser, &warning->location); + // If RubyVM.keep_script_lines is set to true, then we need to create that + // array of script lines here. + if (script_lines != NULL) { + const pm_line_offset_list_t *line_offsets = pm_parser_line_offsets(parser); + *script_lines = rb_ary_new_capa(line_offsets->size); - if (warning->level == PM_WARNING_LEVEL_VERBOSE) { - rb_enc_compile_warning(scope_node->encoding, warning_filepath, line, "%s", warning->message); - } - else { - rb_enc_compile_warn(scope_node->encoding, warning_filepath, line, "%s", warning->message); + for (size_t index = 0; index < line_offsets->size; index++) { + size_t offset = line_offsets->offsets[index]; + size_t length = index == line_offsets->size - 1 ? ((size_t) (pm_parser_end(parser) - (pm_parser_start(parser) + offset))) : (line_offsets->offsets[index + 1] - offset); + rb_ary_push(*script_lines, rb_enc_str_new((const char *) pm_parser_start(parser) + offset, length, scope_node->encoding)); } + + scope_node->script_lines = script_lines; } + // Emit all of the various warnings from the parse. + pm_warning_emit_ctx_t warning_ctx = { + .parser = parser, + .encoding = scope_node->encoding, + .filepath = (const char *) pm_string_source(pm_parser_filepath(parser)) + }; + pm_parser_warnings_each(parser, pm_warning_emit_callback, &warning_ctx); + // If there are errors, raise an appropriate error and free the result. - if (parser->error_list.size > 0) { + if (pm_parser_errors_size(parser) > 0) { VALUE error = pm_parse_process_error(result); // TODO: We need to set the backtrace. @@ -10308,18 +11376,14 @@ pm_parse_process(pm_parse_result_t *result, pm_node_t *node) // Now set up the constant pool and intern all of the various constants into // their corresponding IDs. scope_node->parser = parser; - scope_node->constants = xcalloc(parser->constant_pool.size, sizeof(ID)); - - for (uint32_t index = 0; index < parser->constant_pool.size; index++) { - pm_constant_t *constant = &parser->constant_pool.constants[index]; - scope_node->constants[index] = rb_intern3((const char *) constant->start, constant->length, scope_node->encoding); - } + scope_node->options = result->options; + scope_node->line_offsets = pm_parser_line_offsets(parser); + scope_node->start_line = pm_parser_start_line(parser); + size_t constants_size = pm_parser_constants_size(parser); + scope_node->constants = constants_size ? xmalloc(constants_size * sizeof(ID)) : NULL; - scope_node->index_lookup_table = st_init_numtable(); - pm_constant_id_list_t *locals = &scope_node->locals; - for (size_t index = 0; index < locals->size; index++) { - st_insert(scope_node->index_lookup_table, locals->ids[index], index); - } + pm_intern_constants_ctx_t intern_ctx = { .constants = scope_node->constants, .encoding = scope_node->encoding, .index = 0 }; + pm_parser_constants_each(parser, pm_intern_constants_callback, &intern_ctx); // If we got here, this is a success and we can return Qnil to indicate that // no error should be raised. @@ -10358,22 +11422,22 @@ pm_options_frozen_string_literal_init(pm_options_t *options) static inline VALUE pm_parse_file_script_lines(const pm_scope_node_t *scope_node, const pm_parser_t *parser) { - const pm_newline_list_t *newline_list = &parser->newline_list; - const char *start = (const char *) parser->start; - const char *end = (const char *) parser->end; + const pm_line_offset_list_t *line_offsets = pm_parser_line_offsets(parser); + const char *start = (const char *) pm_parser_start(parser); + const char *end = (const char *) pm_parser_end(parser); // If we end exactly on a newline, then there's no need to push on a final // segment. If we don't, then we need to push on the last offset up to the // end of the string. - size_t last_offset = newline_list->offsets[newline_list->size - 1]; + size_t last_offset = line_offsets->offsets[line_offsets->size - 1]; bool last_push = start + last_offset != end; // Create the ruby strings that represent the lines of the source. - VALUE lines = rb_ary_new_capa(newline_list->size - (last_push ? 0 : 1)); + VALUE lines = rb_ary_new_capa(line_offsets->size - (last_push ? 0 : 1)); - for (size_t index = 0; index < newline_list->size - 1; index++) { - size_t offset = newline_list->offsets[index]; - size_t length = newline_list->offsets[index + 1] - offset; + for (size_t index = 0; index < line_offsets->size - 1; index++) { + size_t offset = line_offsets->offsets[index]; + size_t length = line_offsets->offsets[index + 1] - offset; rb_ary_push(lines, rb_enc_str_new(start + offset, length, scope_node->encoding)); } @@ -10386,6 +11450,37 @@ pm_parse_file_script_lines(const pm_scope_node_t *scope_node, const pm_parser_t return lines; } +struct load_from_fd_args { + VALUE path; + VALUE io; + int open_mode; + int fd; +}; + +static VALUE +close_file(VALUE args) +{ + struct load_from_fd_args *arg = (void *)args; + if (arg->fd != -1) { + close(arg->fd); + } + else if (!NIL_P(arg->io)) { + rb_io_close(arg->io); + } + return Qnil; +} + +static VALUE +load_content(VALUE args) +{ + struct load_from_fd_args *arg = (void *)args; + VALUE io = rb_io_fdopen(arg->fd, arg->open_mode, RSTRING_PTR(arg->path)); + arg->io = io; + arg->fd = -1; + rb_io_wait(io, RB_INT2NUM(RUBY_IO_READABLE), Qnil); + return rb_funcall(io, rb_intern("read"), 0); +} + /** * Attempt to load the file into memory. Return a Ruby error if the file cannot * be read. @@ -10393,32 +11488,67 @@ pm_parse_file_script_lines(const pm_scope_node_t *scope_node, const pm_parser_t VALUE pm_load_file(pm_parse_result_t *result, VALUE filepath, bool load_error) { - if (!pm_string_mapped_init(&result->input, RSTRING_PTR(filepath))) { + pm_source_init_result_t init_result; + result->source = pm_source_mapped_new(RSTRING_PTR(filepath), O_RDONLY | O_NONBLOCK, &init_result); + + if (init_result == PM_SOURCE_INIT_SUCCESS) { + pm_options_frozen_string_literal_init(result->options); + return Qnil; + } + + int err; + + // For non-regular files (pipes, character devices), we need to read + // through Ruby IO to properly release the GVL while waiting for data. + if (init_result == PM_SOURCE_INIT_ERROR_NON_REGULAR) { + struct load_from_fd_args args = { + .path = filepath, + .open_mode = O_RDONLY | O_NONBLOCK, + .fd = rb_cloexec_open(RSTRING_PTR(filepath), args.open_mode, 0), + .io = Qnil, + }; + if (args.fd == -1) goto error_generic; + VALUE contents = rb_ensure(load_content, (VALUE)&args, close_file, (VALUE)&args); + + if (!RB_TYPE_P(contents, T_STRING)) goto error_generic; + + long len = RSTRING_LEN(contents); + if (len < 0) goto error_generic; + + size_t length = (size_t) len; + uint8_t *source_data = xmalloc(length); + memcpy(source_data, RSTRING_PTR(contents), length); + result->source = pm_source_owned_new(source_data, length); + + pm_options_frozen_string_literal_init(result->options); + return Qnil; + } + + if (init_result == PM_SOURCE_INIT_ERROR_DIRECTORY) { + err = EISDIR; + } else { +error_generic: #ifdef _WIN32 - int e = rb_w32_map_errno(GetLastError()); + err = rb_w32_map_errno(GetLastError()); #else - int e = errno; + err = errno; #endif + } - VALUE error; - - if (load_error) { - VALUE message = rb_str_buf_new_cstr(strerror(e)); - rb_str_cat2(message, " -- "); - rb_str_append(message, filepath); + VALUE error; + if (load_error) { + VALUE message = rb_str_buf_new_cstr(strerror(err)); + rb_str_cat2(message, " -- "); + rb_str_append(message, filepath); - error = rb_exc_new3(rb_eLoadError, message); - rb_ivar_set(error, rb_intern_const("@path"), filepath); - } else { - error = rb_syserr_new(e, RSTRING_PTR(filepath)); - RB_GC_GUARD(filepath); - } - - return error; + error = rb_exc_new3(rb_eLoadError, message); + rb_ivar_set(error, rb_intern_const("@path"), filepath); + } else { + error = rb_syserr_new(err, RSTRING_PTR(filepath)); + RB_GC_GUARD(filepath); } - pm_options_frozen_string_literal_init(&result->options); - return Qnil; + return error; } /** @@ -10428,16 +11558,18 @@ pm_load_file(pm_parse_result_t *result, VALUE filepath, bool load_error) * is zeroed out. */ VALUE -pm_parse_file(pm_parse_result_t *result, VALUE filepath) +pm_parse_file(pm_parse_result_t *result, VALUE filepath, VALUE *script_lines) { result->node.filepath_encoding = rb_enc_get(filepath); - pm_options_filepath_set(&result->options, RSTRING_PTR(filepath)); + pm_options_filepath_set(result->options, RSTRING_PTR(filepath)); RB_GC_GUARD(filepath); - pm_parser_init(&result->parser, pm_string_source(&result->input), pm_string_length(&result->input), &result->options); - pm_node_t *node = pm_parse(&result->parser); + pm_options_version_for_current_ruby_set(result->options); + + result->parser = pm_parser_new(result->arena, pm_source_source(result->source), pm_source_length(result->source), result->options); + pm_node_t *node = pm_parse(result->parser); - VALUE error = pm_parse_process(result, node); + VALUE error = pm_parse_process(result, node, script_lines); // If we're parsing a filepath, then we need to potentially support the // SCRIPT_LINES__ constant, which can be a hash that has an array of lines @@ -10445,10 +11577,10 @@ pm_parse_file(pm_parse_result_t *result, VALUE filepath) ID id_script_lines = rb_intern("SCRIPT_LINES__"); if (rb_const_defined_at(rb_cObject, id_script_lines)) { - VALUE script_lines = rb_const_get_at(rb_cObject, id_script_lines); + VALUE constant_script_lines = rb_const_get_at(rb_cObject, id_script_lines); - if (RB_TYPE_P(script_lines, T_HASH)) { - rb_hash_aset(script_lines, filepath, pm_parse_file_script_lines(&result->node, &result->parser)); + if (RB_TYPE_P(constant_script_lines, T_HASH)) { + rb_hash_aset(constant_script_lines, filepath, pm_parse_file_script_lines(&result->node, result->parser)); } } @@ -10460,11 +11592,11 @@ pm_parse_file(pm_parse_result_t *result, VALUE filepath) * cannot be read or if it cannot be parsed properly. */ VALUE -pm_load_parse_file(pm_parse_result_t *result, VALUE filepath) +pm_load_parse_file(pm_parse_result_t *result, VALUE filepath, VALUE *script_lines) { VALUE error = pm_load_file(result, filepath, false); if (NIL_P(error)) { - error = pm_parse_file(result, filepath); + error = pm_parse_file(result, filepath, script_lines); } return error; @@ -10477,27 +11609,43 @@ pm_load_parse_file(pm_parse_result_t *result, VALUE filepath) * error is returned. */ VALUE -pm_parse_string(pm_parse_result_t *result, VALUE source, VALUE filepath) +pm_parse_string(pm_parse_result_t *result, VALUE source, VALUE filepath, VALUE *script_lines) { rb_encoding *encoding = rb_enc_get(source); if (!rb_enc_asciicompat(encoding)) { return rb_exc_new_cstr(rb_eArgError, "invalid source encoding"); } - pm_options_frozen_string_literal_init(&result->options); - pm_string_constant_init(&result->input, RSTRING_PTR(source), RSTRING_LEN(source)); - pm_options_encoding_set(&result->options, rb_enc_name(encoding)); + pm_options_frozen_string_literal_init(result->options); + result->source = pm_source_constant_new((const uint8_t *) RSTRING_PTR(source), (size_t) RSTRING_LEN(source)); + pm_options_encoding_set(result->options, rb_enc_name(encoding)); result->node.filepath_encoding = rb_enc_get(filepath); - pm_options_filepath_set(&result->options, RSTRING_PTR(filepath)); + pm_options_filepath_set(result->options, RSTRING_PTR(filepath)); RB_GC_GUARD(filepath); - pm_parser_init(&result->parser, pm_string_source(&result->input), pm_string_length(&result->input), &result->options); - pm_node_t *node = pm_parse(&result->parser); + pm_options_version_for_current_ruby_set(result->options); - return pm_parse_process(result, node); + result->parser = pm_parser_new(result->arena, pm_source_source(result->source), pm_source_length(result->source), result->options); + pm_node_t *node = pm_parse(result->parser); + + return pm_parse_process(result, node, script_lines); } +struct rb_stdin_wrapper { + VALUE rb_stdin; + int eof_seen; +}; + +static int +pm_parse_stdin_eof(void *stream) +{ + struct rb_stdin_wrapper * wrapped_stdin = (struct rb_stdin_wrapper *)stream; + return wrapped_stdin->eof_seen; +} + +VALUE rb_io_gets_limit_internal(VALUE io, long limit); + /** * An implementation of fgets that is suitable for use with Ruby IO objects. */ @@ -10506,7 +11654,9 @@ pm_parse_stdin_fgets(char *string, int size, void *stream) { RUBY_ASSERT(size > 0); - VALUE line = rb_funcall((VALUE) stream, rb_intern("gets"), 1, INT2FIX(size - 1)); + struct rb_stdin_wrapper * wrapped_stdin = (struct rb_stdin_wrapper *)stream; + + VALUE line = rb_io_gets_limit_internal(wrapped_stdin->rb_stdin, size - 1); if (NIL_P(line)) { return NULL; } @@ -10517,9 +11667,19 @@ pm_parse_stdin_fgets(char *string, int size, void *stream) memcpy(string, cstr, length); string[length] = '\0'; + // We're reading strings from stdin via gets. We'll assume that if the + // string is smaller than the requested length, and doesn't end with a + // newline, that we hit EOF. + if (length < (size - 1) && string[length - 1] != '\n') { + wrapped_stdin->eof_seen = 1; + } + return string; } +// We need access to this function when we're done parsing stdin. +void rb_reset_argf_lineno(long n); + /** * Parse the source off STDIN and store the resulting scope node in the given * parse result struct. It is assumed that the parse result object is zeroed @@ -10528,17 +11688,29 @@ pm_parse_stdin_fgets(char *string, int size, void *stream) VALUE pm_parse_stdin(pm_parse_result_t *result) { - pm_options_frozen_string_literal_init(&result->options); + pm_options_frozen_string_literal_init(result->options); - pm_buffer_t buffer; - pm_node_t *node = pm_parse_stream(&result->parser, &buffer, (void *) rb_stdin, pm_parse_stdin_fgets, &result->options); + struct rb_stdin_wrapper wrapped_stdin = { + rb_stdin, + 0 + }; + + result->source = pm_source_stream_new((void *) &wrapped_stdin, pm_parse_stdin_fgets, pm_parse_stdin_eof); + pm_node_t *node = pm_parse_stream(&result->parser, result->arena, result->source, result->options); + + // When we're done parsing, we reset $. because we don't want the fact that + // we went through an IO object to be visible to the user. + rb_reset_argf_lineno(0); + + return pm_parse_process(result, node, NULL); +} - // Copy the allocated buffer contents into the input string so that it gets - // freed. At this point we've handed over ownership, so we don't need to - // free the buffer itself. - pm_string_owned_init(&result->input, (uint8_t *) pm_buffer_value(&buffer), pm_buffer_length(&buffer)); +#define PM_VERSION_FOR_RELEASE(major, minor) PM_VERSION_FOR_RELEASE_IMPL(major, minor) +#define PM_VERSION_FOR_RELEASE_IMPL(major, minor) #major "." #minor - return pm_parse_process(result, node); +void pm_options_version_for_current_ruby_set(pm_options_t *options) { + const char *version = PM_VERSION_FOR_RELEASE(RUBY_API_VERSION_MAJOR, RUBY_API_VERSION_MINOR); + pm_options_version_set(options, version, strlen(version)); } #undef NEW_ISEQ |