Re-apply "Ruby 3.2 - Speed up rebuilding the loaded feature index and realpath cache (#8023)" (#8252)

* Re-apply "Ruby 3.2 - Speed up rebuilding the loaded feature index and realpath cache (#8023)"

* [CI] mingw.yml - remove IBM437 encoding for test-all, use cmd shell for test & test-all

* Skip failing test on mingw with readline.so

Co-authored-by: nagachika <nagachika@ruby-lang.org>

---------

Co-authored-by: nagachika <nagachika@ruby-lang.org>

5 files changed, 50 insertions, 8 deletions

diff --git a/.github/workflows/mingw.yml b/.github/workflows/mingw.yml
index ba3886218a..c8b53ad76b 100644
--- a/.github/workflows/mingw.yml
+++ b/.github/workflows/mingw.yml
@@ -139,12 +139,11 @@ jobs:
         run: |
           make test
         if: ${{matrix.test_task == 'check' || matrix.test_task == 'test'}}
+        shell: cmd
 
       - name: test-all
         timeout-minutes: 45
         run: |
-          # Actions uses UTF8, causes test failures, similar to normal OS setup
-          chcp.com 437
           make ${{ StartsWith(matrix.test_task, 'test/') && matrix.test_task || 'test-all' }}
         env:
           RUBY_TESTOPTS: >-
@@ -152,6 +151,7 @@ jobs:
             ${{ matrix.test-all-opts }}
           BUNDLER_VERSION:
         if: ${{matrix.test_task == 'check' || matrix.test_task == 'test-all' || StartsWith(matrix.test_task, 'test/')}}
+        shell: cmd
 
       - name: test-spec
         timeout-minutes: 10
diff --git a/load.c b/load.c
index f3d39c2c08..44449fc77e 100644
--- a/load.c
+++ b/load.c
@@ -164,6 +164,12 @@ get_loaded_features_realpaths(rb_vm_t *vm)
 }
 
 static VALUE
+get_loaded_features_realpath_map(rb_vm_t *vm)
+{
+    return vm->loaded_features_realpath_map;
+}
+
+static VALUE
 get_LOADED_FEATURES(ID _x, VALUE *_y)
 {
     return get_loaded_features(GET_VM());
@@ -361,7 +367,10 @@ get_loaded_features_index(rb_vm_t *vm)
         st_foreach(vm->loaded_features_index, loaded_features_index_clear_i, 0);
 
         VALUE realpaths = vm->loaded_features_realpaths;
+        VALUE realpath_map = vm->loaded_features_realpath_map;
+        VALUE previous_realpath_map = rb_hash_dup(realpath_map);
         rb_hash_clear(realpaths);
+        rb_hash_clear(realpath_map);
         features = vm->loaded_features;
         for (i = 0; i < RARRAY_LEN(features); i++) {
             VALUE entry, as_str;
@@ -378,9 +387,14 @@ get_loaded_features_index(rb_vm_t *vm)
         long j = RARRAY_LEN(features);
         for (i = 0; i < j; i++) {
             VALUE as_str = rb_ary_entry(features, i);
-            VALUE realpath = rb_check_realpath(Qnil, as_str, NULL);
-            if (NIL_P(realpath)) realpath = as_str;
-            rb_hash_aset(realpaths, rb_fstring(realpath), Qtrue);
+            VALUE realpath = rb_hash_aref(previous_realpath_map, as_str);
+            if (NIL_P(realpath)) {
+                realpath = rb_check_realpath(Qnil, as_str, NULL);
+                if (NIL_P(realpath)) realpath = as_str;
+                realpath = rb_fstring(realpath);
+            }
+            rb_hash_aset(realpaths, realpath, Qtrue);
+            rb_hash_aset(realpath_map, as_str, realpath);
         }
     }
     return vm->loaded_features_index;
@@ -675,6 +689,19 @@ rb_provide(const char *feature)
 
 NORETURN(static void load_failed(VALUE));
 
+static inline VALUE
+realpath_internal_cached(VALUE hash, VALUE path)
+{
+    VALUE ret = rb_hash_aref(hash, path);
+    if(RTEST(ret)) {
+        return ret;
+    }
+
+    VALUE realpath = rb_realpath_internal(Qnil, path, 1);
+    rb_hash_aset(hash, rb_fstring(path), rb_fstring(realpath));
+    return realpath;
+}
+
 static inline void
 load_iseq_eval(rb_execution_context_t *ec, VALUE fname)
 {
@@ -687,8 +714,12 @@ load_iseq_eval(rb_execution_context_t *ec, VALUE fname)
         VALUE parser = rb_parser_new();
         rb_parser_set_context(parser, NULL, FALSE);
         ast = (rb_ast_t *)rb_parser_load_file(parser, fname);
+
+        rb_thread_t *th = rb_ec_thread_ptr(ec);
+        VALUE realpath_map = get_loaded_features_realpath_map(th->vm);
+
         iseq = rb_iseq_new_top(&ast->body, rb_fstring_lit("<top (required)>"),
-                               fname, rb_realpath_internal(Qnil, fname, 1), NULL);
+                               fname, realpath_internal_cached(realpath_map, fname), NULL);
         rb_ast_dispose(ast);
         rb_vm_pop_frame(ec);
         RB_GC_GUARD(v);
@@ -1161,6 +1192,7 @@ require_internal(rb_execution_context_t *ec, VALUE fname, int exception, bool wa
     volatile VALUE saved_path;
     volatile VALUE realpath = 0;
     VALUE realpaths = get_loaded_features_realpaths(th->vm);
+    VALUE realpath_map = get_loaded_features_realpath_map(th->vm);
     volatile bool reset_ext_config = false;
     struct rb_ext_config prev_ext_config;
 
@@ -1194,7 +1226,7 @@ require_internal(rb_execution_context_t *ec, VALUE fname, int exception, bool wa
             }
 #endif
             else if (RTEST(rb_hash_aref(realpaths,
-                                        realpath = rb_realpath_internal(Qnil, path, 1)))) {
+                                        realpath = realpath_internal_cached(realpath_map, path)))) {
                 result = 0;
             }
             else {
@@ -1252,7 +1284,8 @@ require_internal(rb_execution_context_t *ec, VALUE fname, int exception, bool wa
         rb_provide_feature(th2->vm, path);
         VALUE real = realpath;
         if (real) {
-            rb_hash_aset(realpaths, rb_fstring(real), Qtrue);
+            real = rb_fstring(real);
+            rb_hash_aset(realpaths, real, Qtrue);
         }
     }
     ec->errinfo = saved.errinfo;
@@ -1473,6 +1506,8 @@ Init_load(void)
     vm->loaded_features_index = st_init_numtable();
     vm->loaded_features_realpaths = rb_hash_new();
     rb_obj_hide(vm->loaded_features_realpaths);
+    vm->loaded_features_realpath_map = rb_hash_new();
+    rb_obj_hide(vm->loaded_features_realpath_map);
 
     rb_define_global_function("load", rb_f_load, -1);
     rb_define_global_function("require", rb_f_require, 1);
diff --git a/test/excludes/TestReadline.rb b/test/excludes/TestReadline.rb
new file mode 100644
index 0000000000..706736f6f6
--- /dev/null
+++ b/test/excludes/TestReadline.rb
@@ -0,0 +1,4 @@
+# frozen_string_literal: false
+if /mingw/i.match?(RUBY_PLATFORM)
+  exclude(:test_input_metachar_multibyte, "failed with readline.so on MiNGW")
+end
diff --git a/vm.c b/vm.c
index d009a5f64a..de43d022c0 100644
--- a/vm.c
+++ b/vm.c
@@ -2703,6 +2703,7 @@ rb_vm_update_references(void *ptr)
         vm->loaded_features = rb_gc_location(vm->loaded_features);
         vm->loaded_features_snapshot = rb_gc_location(vm->loaded_features_snapshot);
         vm->loaded_features_realpaths = rb_gc_location(vm->loaded_features_realpaths);
+        vm->loaded_features_realpath_map = rb_gc_location(vm->loaded_features_realpath_map);
         vm->top_self = rb_gc_location(vm->top_self);
         vm->orig_progname = rb_gc_location(vm->orig_progname);
 
@@ -2794,6 +2795,7 @@ rb_vm_mark(void *ptr)
         rb_gc_mark_movable(vm->loaded_features);
         rb_gc_mark_movable(vm->loaded_features_snapshot);
         rb_gc_mark_movable(vm->loaded_features_realpaths);
+        rb_gc_mark_movable(vm->loaded_features_realpath_map);
         rb_gc_mark_movable(vm->top_self);
         rb_gc_mark_movable(vm->orig_progname);
         RUBY_MARK_MOVABLE_UNLESS_NULL(vm->coverages);
diff --git a/vm_core.h b/vm_core.h
index d86fdbaecd..b6adeadd87 100644
--- a/vm_core.h
+++ b/vm_core.h
@@ -680,6 +680,7 @@ typedef struct rb_vm_struct {
     VALUE loaded_features;
     VALUE loaded_features_snapshot;
     VALUE loaded_features_realpaths;
+    VALUE loaded_features_realpath_map;
     struct st_table *loaded_features_index;
     struct st_table *loading_table;
 #if EXTSTATIC