1 files changed, 81 insertions, 10 deletions

diff --git a/include/ruby/internal/core/rmatch.h b/include/ruby/internal/core/rmatch.h
index 03ab5e5d82..a528c2999e 100644
--- a/include/ruby/internal/core/rmatch.h
+++ b/include/ruby/internal/core/rmatch.h
@@ -17,18 +17,23 @@
  *             recursively included  from extension  libraries written  in C++.
  *             Do not  expect for  instance `__VA_ARGS__` is  always available.
  *             We assume C99  for ruby itself but we don't  assume languages of
- *             extension libraries. They could be written in C++98.
+ *             extension libraries.  They could be written in C++98.
  * @brief      Defines struct ::RMatch.
  */
 #include "ruby/internal/attr/artificial.h"
 #include "ruby/internal/attr/pure.h"
-#include "ruby/internal/attr/returns_nonnull.h"
 #include "ruby/internal/cast.h"
 #include "ruby/internal/core/rbasic.h"
 #include "ruby/internal/value.h"
 #include "ruby/internal/value_type.h"
 #include "ruby/assert.h"
 
+/**
+ * Convenient casting macro.
+ *
+ * @param   obj  An object, which is in fact an ::RMatch.
+ * @return  The passed object casted to ::RMatch.
+ */
 #define RMATCH(obj) RBIMPL_CAST((struct RMatch *)(obj))
 /** @cond INTERNAL_MACRO */
 #define RMATCH_REGS RMATCH_REGS
@@ -37,37 +42,103 @@
 struct re_patter_buffer; /* a.k.a. OnigRegexType, defined in onigmo.h */
 struct re_registers;     /* Also in onigmo.h */
 
-/* @shyouhei wonders: is anyone actively using this typedef ...? */
+/**
+ * @old{re_pattern_buffer}
+ *
+ * @internal
+ *
+ * @shyouhei wonders: is anyone actively using this typedef ...?
+ */
 typedef struct re_pattern_buffer Regexp;
 
+/**
+ * Represents the  region of a  capture group.   This is basically  for caching
+ * purpose.  re_registers have similar concepts  (`beg` and `end`) but they are
+ * in `ptrdiff_t*`.  In order for  us to implement `MatchData#offset` that info
+ * has to  be converted to  offset integers.  This is  the struct to  hold such
+ * things.
+ *
+ * @internal
+ *
+ * But why on earth it has to be visible from extension libraries?
+ */
 struct rmatch_offset {
-    long beg;
-    long end;
+    long beg; /**< Beginning of a group. */
+    long end; /**< End of a group. */
 };
 
-struct rmatch {
+/** Represents a match. */
+struct rb_matchext_struct {
+    /**
+     * "Registers"  of a  match.   This  is a  quasi-opaque  struct that  holds
+     * execution result of a match.  Roughly resembles `&~`.
+     */
     struct re_registers regs;
 
+    /** Capture group offsets, in C array. */
     struct rmatch_offset *char_offset;
+
+    /** Number of ::rmatch_offset that ::rmatch::char_offset holds. */
     int char_offset_num_allocated;
 };
 
+typedef struct rb_matchext_struct rb_matchext_t;
+
+/**
+ * Regular expression  execution context.  When a  regular expression "matches"
+ * to a string, it generates capture  groups etc.  This struct holds that info.
+ * Visible from Ruby as an instance of `MatchData`.
+ *
+ * @note  There is  no way  for extension libraries  to manually  generate this
+ *        struct except by actually exercising the match operation of a regular
+ *        expression.
+ */
 struct RMatch {
+
+    /** Basic part, including flags and class. */
     struct RBasic basic;
+
+    /**
+     * The target string that the match was made against.
+     */
     VALUE str;
-    struct rmatch *rmatch;
+
+    /**
+     * The expression of this match.
+     */
     VALUE regexp;  /* RRegexp */
 };
 
+#define RMATCH_EXT(m) ((rb_matchext_t *)((char *)(m) + sizeof(struct RMatch)))
+
 RBIMPL_ATTR_PURE_UNLESS_DEBUG()
-RBIMPL_ATTR_RETURNS_NONNULL()
 RBIMPL_ATTR_ARTIFICIAL()
+/**
+ * Queries the raw ::re_registers.
+ *
+ * @param[in]  match  A match object
+ * @pre        `match` must be of ::RMatch.
+ * @return     Its execution result.
+ * @note       Good.  So you  are aware of the fact that  it could return NULL.
+ *             Yes.  It  actually does.  This  is a really bizarre  thing.  The
+ *             situation  is about  `String#gsub`  and its  family.  They  take
+ *             strings as  arguments, like `"foo".sub("bar", "baz")`.   On such
+ *             situations,  in  order  to optimise  memory  allocations,  these
+ *             methods do  not involve regular  expressions at all.   They just
+ *             sequentially scan  the receiver.  Okay.  The  story begins here.
+ *             Even when  they do  not kick  our regexp  engine, there  must be
+ *             backref objects e.g. `$&`.  But how?  You know what?  Ruby fakes
+ *             them.  It  allocates an empty  ::RMatch and behaves as  if there
+ *             were  execution   contexts.   In  reality  there   weren't.   No
+ *             ::re_registers are  allocated then.   There is  no way  for this
+ *             function but  to return NULL  for those fake ::RMatch.   This is
+ *             the reason for the nullability of this function.
+ */
 static inline struct re_registers *
 RMATCH_REGS(VALUE match)
 {
     RBIMPL_ASSERT_TYPE(match, RUBY_T_MATCH);
-    RBIMPL_ASSERT_OR_ASSUME(RMATCH(match)->rmatch != NULL);
-    return &RMATCH(match)->rmatch->regs;
+    return &RMATCH_EXT(match)->regs;
 }
 
 #endif /* RBIMPL_RMATCH_H */