From b40d74ce696ca3c58e051220fafdaeb2d2e20214 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E5=8D=9C=E9=83=A8=E6=98=8C=E5=B9=B3?= Date: Mon, 1 Feb 2021 16:52:16 +0900 Subject: include/ruby/internal/core/rmatch.h: add doxygen Must not be a bad idea to improve documents. [ci skip] --- include/ruby/internal/core/rmatch.h | 83 ++++++++++++++++++++++++++++++++++--- 1 file changed, 78 insertions(+), 5 deletions(-) diff --git a/include/ruby/internal/core/rmatch.h b/include/ruby/internal/core/rmatch.h index 0db21b0088..2d2fd897f5 100644 --- a/include/ruby/internal/core/rmatch.h +++ b/include/ruby/internal/core/rmatch.h @@ -22,13 +22,18 @@ */ #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 @@ -37,31 +42,99 @@ 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. */ }; +/** Represents a match. */ struct rmatch { + /** + * "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; }; +/** + * 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; + + /** + * The result of this match. + */ struct rmatch *rmatch; + + /** + * The expression of this match. + */ VALUE regexp; /* RRegexp */ }; 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) { -- cgit v1.2.3