diff --git a/include/ruby/internal/intern/gc.h b/include/ruby/internal/intern/gc.h index a178808a52..1617a7cef6 100644 --- a/include/ruby/internal/intern/gc.h +++ b/include/ruby/internal/intern/gc.h @@ -20,37 +20,370 @@ * extension libraries. They could be written in C++98. * @brief Public APIs related to ::rb_mGC. */ +#include "ruby/internal/config.h" + +#ifdef STDC_HEADERS +# include /* size_t */ +#endif + +#if HAVE_SYS_TYPES_H +# include /* ssize_t */ +#endif + +#include "ruby/internal/attr/cold.h" +#include "ruby/internal/attr/noreturn.h" +#include "ruby/internal/attr/nonnull.h" +#include "ruby/internal/attr/pure.h" #include "ruby/internal/dllexport.h" #include "ruby/internal/value.h" -#include "ruby/backward/2/attributes.h" RBIMPL_SYMBOL_EXPORT_BEGIN() /* gc.c */ -COLDFUNC NORETURN(void rb_memerror(void)); -PUREFUNC(int rb_during_gc(void)); -void rb_gc_mark_locations(const VALUE*, const VALUE*); -void rb_mark_tbl(struct st_table*); -void rb_mark_tbl_no_pin(struct st_table*); -void rb_mark_set(struct st_table*); -void rb_mark_hash(struct st_table*); + +RBIMPL_ATTR_COLD() +RBIMPL_ATTR_NORETURN() +/** + * Triggers out-of-memory error. If possible it raises ::rb_eNoMemError. But + * because we are running out of memory that is not always doable. This + * function tries hard to show something, but ultimately can die silently. + * + * @exception rb_eNoMemError Raises it if possible. + */ +void rb_memerror(void); + +RBIMPL_ATTR_PURE() +/** + * Queries if the GC is busy. + * + * @retval 0 It isn't. + * @retval 1 It is. + */ +int rb_during_gc(void); + +RBIMPL_ATTR_NONNULL((1)) +/** + * Marks objects between the two pointers. This is one of the GC utility + * functions that you can call when you design your own + * ::rb_data_type_struct::dmark. + * + * @pre Continuous memory region from `start` to `end` shall be fully + * addressable. + * @param[out] start Pointer to an array of objects. + * @param[out] end Pointer that terminates the array of objects. + * @post Objects from `start` to `end`, both inclusive, are marked. + * + * @internal + * + * `end` can be NULL... But that just results in no-op. + */ +void rb_gc_mark_locations(const VALUE *start, const VALUE *end); + +/** + * Identical to rb_mark_hash(), except it marks only values of the table and + * leave their associated keys unmarked. This is one of the GC utility + * functions that you can call when you design your own + * ::rb_data_type_struct::dmark. + * + * @warning Of course it can break GC. Leave it unused if unsure. + * @param[in] tbl A table to mark. + * @post Values stored in `tbl` are marked. + */ +void rb_mark_tbl(struct st_table *tbl); + +/** + * Identical to rb_mark_tbl(), except it marks objects using + * rb_gc_mark_movable(). This is one of the GC utility functions that you can + * call when you design your own ::rb_data_type_struct::dmark. + * + * @warning Of course it can break GC. Leave it unused if unsure. + * @param[in] tbl A table to mark. + * @post Values stored in `tbl` are marked. + */ +void rb_mark_tbl_no_pin(struct st_table *tbl); + +/** + * Identical to rb_mark_hash(), except it marks only keys of the table and + * leave their associated values unmarked. This is one of the GC utility + * functions that you can call when you design your own + * ::rb_data_type_struct::dmark. + * + * @warning Of course it can break GC. Leave it unused if unsure. + * @param[in] tbl A table to mark. + * @post Keys stored in `tbl` are marked. + */ +void rb_mark_set(struct st_table *tbl); + +/** + * Marks keys and values associated inside of the given table. This is one of + * the GC utility functions that you can call when you design your own + * ::rb_data_type_struct::dmark. + * + * @param[in] tbl A table to mark. + * @post Objects stored in `tbl` are marked. + */ +void rb_mark_hash(struct st_table *tbl); + +/** + * Updates references inside of tables. After you marked values using + * rb_mark_tbl_no_pin(), the objects inside of the table could of course be + * moved. This function is to fixup those references. You can call this from + * your ::rb_data_type_struct::dcompact. + * + * @param[out] ptr A table that potentially includes moved references. + * @post Moved references, if any, are corrected. + */ void rb_gc_update_tbl_refs(st_table *ptr); -void rb_gc_mark_maybe(VALUE); -void rb_gc_mark(VALUE); -void rb_gc_mark_movable(VALUE); -VALUE rb_gc_location(VALUE); -void rb_gc_force_recycle(VALUE); + +/** + * Identical to rb_gc_mark(), except it allows the passed value be a + * non-object. For instance pointers to different type of memory regions are + * allowed here. Such values are silently ignored. This is one of the GC + * utility functions that you can call when you design your own + * ::rb_data_type_struct::dmark. + * + * @param[out] obj A possible object. + * @post `obj` is marked, if possible. + */ +void rb_gc_mark_maybe(VALUE obj); + +/** + * Marks an object. This is one of the GC utility functions that you can call + * when you design your own ::rb_data_type_struct::dmark. + * + * @param[out] obj Arbitrary Ruby object. + * @post `obj` is marked. + */ +void rb_gc_mark(VALUE obj); + +/** + * Maybe this is the only function provided for C extensions to control the + * pinning of objects, so let us describe it in detail. These days Ruby's GC + * is copying. As far as an object's physical address is guaranteed unused, it + * can move around the object space. Our GC engine rearranges these objects + * after it reclaims unreachable objects from our object space, so that the + * space is compact (improves memory locality). This is called the + * "compaction" phase, and works well most of the time... as far as there are + * no C extensions. C extensions complicate the scenario because Ruby core + * cannot detect any use of the physical address of an object inside of C + * functions. In order to prevent memory corruptions, objects observable from + * C extensions are "pinned"; they stick to where they are born until they die, + * just in case any C extensions touch their raw pointers. This variant of + * scheme is called "Mostly-Copying" garbage collector. Authors of C + * extensions, however, can extremely carefully write them to become + * compaction-aware. To do so avoid referring to a Ruby object from inside of + * your struct in the first place. But if that is not possible, use this + * function from your ::rb_data_type_struct::dmark then. This way objects + * marked using it are considered movable. If you chose this way you have to + * manually fix up locations of such moved pointers using rb_gc_location(). + * + * @see Bartlett, Joel F., "Compacting Garbage Collection with Ambiguous + * Roots", ACM SIGPLAN Lisp Pointers Volume 1 Issue 6 pp. 3-12, + * April-May-June, 1988. https://doi.org/10.1145/1317224.1317225 + * + * @param[in] obj Object that is movable. + * @post Values stored in `tbl` are marked. + */ +void rb_gc_mark_movable(VALUE obj); + +/** + * Finds a new "location" of an object. An object can be moved on compaction. + * This function projects its new abode, or just returns the passed object if + * not moved. This is one of the GC utility functions that you can call when + * you design your own ::rb_data_type_struct::dcompact. + * + * @param[in] obj An object, possibly already moved to somewhere else. + * @return An object, which holds the current contents of former `obj`. + */ +VALUE rb_gc_location(VALUE obj); + +/** + * Asserts that the passed object is no longer needed. Such objects are + * reclaimed sooner or later so this function is not mandatory. But sometimes + * you can know from your application knowledge that an object is surely dead + * at some point. Calling this as a hint can be a polite way. + * + * @param[out] obj Object, dead. + * @pre `obj` have never been passed to this function before. + * @post `obj` could be invalidated. + * @warning It is a failure to pass an object multiple times to this + * function. + */ +void rb_gc_force_recycle(VALUE obj); + +/** + * Triggers a GC process. This was the only GC entry point that we had at the + * beginning. Over time our GC evolved. Now what this function does is just a + * very simplified variation of the entire GC algorithms. A series of + * procedures kicked by this API is called a "full" GC. + * + * - It immediately scans the entire object space to sort the dead. + * - It immediately reclaims any single dead bodies to reuse later. + * + * It is worth noting that the procedures above do not include evaluations of + * finalisers. They run later. + * + * @internal + * + * Finalisers are deferred until we can handle interrupts. See + * `rb_postponed_job_flush` in vm_trace.c. + * + * Of course there are GC that are not "full". For instance this one and the + * GC which runs when we are running out of memory are different. See + * `gc_profile_record_flag` defined in gc.c for the kinds of GC. + * + * In spite of the name this is not what everything that a GC can trigger. As + * of writing it seems this function does not trigger compaction. But this + * might change in future. + */ void rb_gc(void); -void rb_gc_copy_finalizer(VALUE,VALUE); + +/** + * Copy&paste an object's finaliser to another. This is one of the GC utility + * functions that you can call when you design your own `initialize_copy`, + * `initialize_dup`, `initialize_clone`. + * + * @param[out] dst Destination object. + * @param[in] src Source object. + * @post `dst` and `src` share the same finaliser. + * + * @internal + * + * But isn't it easier for you to call super, and let `Object#intialize_copy` + * call this function instead? + */ +void rb_gc_copy_finalizer(VALUE dst, VALUE src); + +/** + * (Re-) enables GC. This makes sense only after you called rb_gc_disable(). + * + * @retval RUBY_Qtrue GC was disabled before. + * @retval RUBY_Qfalse GC was enabled before. + * @post GC is enabled. + * + * @internal + * + * This is one of such exceptional functions that does not raise both Ruby + * exceptions and C++ exceptions. + */ VALUE rb_gc_enable(void); + +/** + * Disables GC. This prevents automatic GC runs when the process is running + * out of memory. Such situations shall result in rb_memerror(). However this + * does not prevent users from manually invoking rb_gc(). That should work. + * People abused this by disabling GC at the beginning of an event loop, + * process events without GC overheads, then manually force reclaiming garbage + * at the bottom of the loop. However because our GC is now much smarter than + * just calling rb_gc(), this technique is proven to be sub-optimal these days. + * It is believed that there is currently practically no needs of this + * function. + * + * @retval RUBY_Qtrue GC was disabled before. + * @retval RUBY_Qfalse GC was enabled before. + * @post GC is disabled. + */ VALUE rb_gc_disable(void); + +/** + * Identical to rb_gc(), except the return value. + * + * @return Always returns ::RUBY_Qnil. + */ VALUE rb_gc_start(void); -VALUE rb_define_finalizer(VALUE, VALUE); -VALUE rb_undefine_finalizer(VALUE); + +/** + * Assigns a finaliser for an object. Each objects can have objects (typically + * blocks) that run immediately after that object dies. They are called + * finalisers of an object. This function associates a finaliser object with a + * target object. + * + * @note Note that finalisers run _after_ the object they finalise dies. You + * cannot for instance call its methods. + * @note If your finaliser references the object it finalises that object + * loses any chance to become a garbage; effectively leaks memory until + * the end of the process. + * + * @param[in] obj Target to finalise. + * @param[in] block Something `call`able. + * @exception rb_eRuntimeError Somehow `obj` cannot have finalisers. + * @exception rb_eFrozenError `obj` is frozen. + * @exception rb_eArgError `block` doesn't respond to `call`. + * @return The passed `block`. + * @post `block` runs after `obj` dies. + */ +VALUE rb_define_finalizer(VALUE obj, VALUE block); + +/** + * Modifies the object so that it has no finalisers at all. This function is + * mainly provided for symmetry. No practical usages can be thought of. + * + * @param[out] obj Object to clear its finalisers. + * @exception rb_eFrozenError `obj` is frozen. + * @return The passed `obj`. + * @post `obj` has no finalisers. + * @note There is no way to undefine a specific part of many finalisers + * that `obj` could have. All you can do is to clear them all. + */ +VALUE rb_undefine_finalizer(VALUE obj); + +/** + * Identical to rb_gc_stat(), with "count" parameter. + * + * @return Lifetime total number of runs of GC. + */ size_t rb_gc_count(void); -size_t rb_gc_stat(VALUE); -VALUE rb_gc_latest_gc_info(VALUE); -void rb_gc_adjust_memory_usage(ssize_t); + +/** + * Obtains various GC related profiles. The parameter can be either a Symbol + * or a Hash. If a Hash is passed, it is filled with everything currently + * available. If a Symbol is passed just that portion is returned. + * + * Possible variations of keys you can pass here change from version to + * version. You can get the list of known keys by passing an empty hash and + * let it be filled. + * + * @param[in,out] key_or_buf A Symbol, or a Hash. + * @exception rb_eTypeError Neither Symbol nor Hash. + * @exception rb_eFrozenError Frozen hash is passed. + * @return In case a Hash is passed it returns 0. Otherwise the + * profile value associated with the given key is returned. + * @post In case a Hash is passed it is filled with values. + */ +size_t rb_gc_stat(VALUE key_or_buf); + +/** + * Obtains various info regarding the most recent GC run. This includes for + * instance the reason of the GC. The parameter can be either a Symbol or a + * Hash. If a Hash is passed, it is filled with everything currently + * available. If a Symbol is passed just that portion is returned. + * + * Possible variations of keys you can pass here change from version to + * version. You can get the list of known keys by passing an empty hash and + * let it be filled. + * + * @param[in,out] key_or_buf A Symbol, or a Hash. + * @exception rb_eTypeError Neither Symbol nor Hash. + * @exception rb_eFrozenError Frozen hash is passed. + * @return In case a Hash is passed it returns that hash. Otherwise + * the profile value associated with the given key is returned. + * @post In case a Hash is passed it is filled with values. + */ +VALUE rb_gc_latest_gc_info(VALUE key_or_buf); + +/** + * Informs that there are external memory usages. Our GC runs when we are + * running out of memory. The amount of memory, however, can increase/decrease + * behind-the-scene. For instance DLLs can allocate memories using `mmap(2)` + * etc, which are opaque to us. Registering such external allocations using + * this function enables proper detection of how much memories an object used + * as a whole. That will trigger GCs more often than it would otherwise. You + * can also pass negative numbers here, to indicate that such external + * allocations are gone. + * + * @param[in] diff Amount of memory increased(+)/decreased(-). + */ +void rb_gc_adjust_memory_usage(ssize_t diff); RBIMPL_SYMBOL_EXPORT_END()