ruby/variable.c

3806 строки
88 KiB
C
Исходник Обычный вид История

/**********************************************************************
variable.c -
$Author$
created at: Tue Apr 19 23:55:15 JST 1994
* encoding.c: provide basic features for M17N. * parse.y: encoding aware parsing. * parse.y (pragma_encoding): encoding specification pragma. * parse.y (rb_intern3): encoding specified symbols. * string.c (rb_str_length): length based on characters. for older behavior, bytesize method added. * string.c (rb_str_index_m): index based on characters. rindex as well. * string.c (succ_char): encoding aware succeeding string. * string.c (rb_str_reverse): reverse based on characters. * string.c (rb_str_inspect): encoding aware string description. * string.c (rb_str_upcase_bang): encoding aware case conversion. downcase, capitalize, swapcase as well. * string.c (rb_str_tr_bang): tr based on characters. delete, squeeze, tr_s, count as well. * string.c (rb_str_split_m): split based on characters. * string.c (rb_str_each_line): encoding aware each_line. * string.c (rb_str_each_char): added. iteration based on characters. * string.c (rb_str_strip_bang): encoding aware whitespace stripping. lstrip, rstrip as well. * string.c (rb_str_justify): encoding aware justifying (ljust, rjust, center). * string.c (str_encoding): get encoding attribute from a string. * re.c (rb_reg_initialize): encoding aware regular expression * sprintf.c (rb_str_format): formatting (i.e. length count) based on characters. * io.c (rb_io_getc): getc to return one-character string. for older behavior, getbyte method added. * ext/stringio/stringio.c (strio_getc): ditto. * io.c (rb_io_ungetc): allow pushing arbitrary string at the current reading point. * ext/stringio/stringio.c (strio_ungetc): ditto. * ext/strscan/strscan.c: encoding support. git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@13261 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2007-08-25 07:29:39 +04:00
Copyright (C) 1993-2007 Yukihiro Matsumoto
Copyright (C) 2000 Network Applied Communication Laboratory, Inc.
Copyright (C) 2000 Information-technology Promotion Agency, Japan
**********************************************************************/
#include "ruby/internal/config.h"
#include <stddef.h>
#include "ruby/internal/stdbool.h"
#include "ccan/list/list.h"
#include "constant.h"
#include "debug_counter.h"
#include "id.h"
#include "id_table.h"
#include "internal.h"
#include "internal/class.h"
#include "internal/compilers.h"
#include "internal/error.h"
#include "internal/eval.h"
#include "internal/hash.h"
#include "internal/object.h"
#include "internal/re.h"
#include "internal/symbol.h"
#include "internal/thread.h"
#include "internal/variable.h"
#include "ruby/encoding.h"
#include "ruby/st.h"
#include "ruby/util.h"
#include "transient_heap.h"
#include "variable.h"
#include "vm_core.h"
#include "ractor_core.h"
#include "vm_sync.h"
Add a cache for class variables Redo of 34a2acdac788602c14bf05fb616215187badd504 and 931138b00696419945dc03e10f033b1f53cd50f3 which were reverted. GitHub PR #4340. This change implements a cache for class variables. Previously there was no cache for cvars. Cvar access is slow due to needing to travel all the way up th ancestor tree before returning the cvar value. The deeper the ancestor tree the slower cvar access will be. The benefits of the cache are more visible with a higher number of included modules due to the way Ruby looks up class variables. The benchmark here includes 26 modules and shows with the cache, this branch is 6.5x faster when accessing class variables. ``` compare-ruby: ruby 3.1.0dev (2021-03-15T06:22:34Z master 9e5105c) [x86_64-darwin19] built-ruby: ruby 3.1.0dev (2021-03-15T12:12:44Z add-cache-for-clas.. c6be009) [x86_64-darwin19] | |compare-ruby|built-ruby| |:--------|-----------:|---------:| |vm_cvar | 5.681M| 36.980M| | | -| 6.51x| ``` Benchmark.ips calling `ActiveRecord::Base.logger` from within a Rails application. ActiveRecord::Base.logger has 71 ancestors. The more ancestors a tree has, the more clear the speed increase. IE if Base had only one ancestor we'd see no improvement. This benchmark is run on a vanilla Rails application. Benchmark code: ```ruby require "benchmark/ips" require_relative "config/environment" Benchmark.ips do |x| x.report "logger" do ActiveRecord::Base.logger end end ``` Ruby 3.0 master / Rails 6.1: ``` Warming up -------------------------------------- logger 155.251k i/100ms Calculating ------------------------------------- ``` Ruby 3.0 with cvar cache / Rails 6.1: ``` Warming up -------------------------------------- logger 1.546M i/100ms Calculating ------------------------------------- logger 14.857M (± 4.8%) i/s - 74.198M in 5.006202s ``` Lastly we ran a benchmark to demonstate the difference between master and our cache when the number of modules increases. This benchmark measures 1 ancestor, 30 ancestors, and 100 ancestors. Ruby 3.0 master: ``` Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower ``` Ruby 3.0 with cvar cache: ``` Warming up -------------------------------------- 1 module 1.641M i/100ms 30 modules 1.655M i/100ms 100 modules 1.620M i/100ms Calculating ------------------------------------- 1 module 16.279M (± 3.8%) i/s - 82.038M in 5.046923s 30 modules 15.891M (± 3.9%) i/s - 79.459M in 5.007958s 100 modules 16.087M (± 3.6%) i/s - 81.005M in 5.041931s Comparison: 1 module: 16279458.0 i/s 100 modules: 16087484.6 i/s - same-ish: difference falls within error 30 modules: 15891406.2 i/s - same-ish: difference falls within error ``` Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
2021-06-01 20:34:06 +03:00
RUBY_EXTERN rb_serial_t ruby_vm_global_cvar_state;
#define GET_GLOBAL_CVAR_STATE() (ruby_vm_global_cvar_state)
typedef void rb_gvar_compact_t(void *var);
static struct rb_id_table *rb_global_tbl;
static ID autoload, classpath, tmp_classpath;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
static VALUE autoload_featuremap; /* feature => autoload_i */
static void check_before_mod_set(VALUE, ID, VALUE, const char *);
static void setup_const_entry(rb_const_entry_t *, VALUE, VALUE, rb_const_flag_t);
static VALUE rb_const_search(VALUE klass, ID id, int exclude, int recurse, int visibility);
static st_table *generic_iv_tbl_;
struct ivar_update {
union {
st_table *iv_index_tbl;
struct gen_ivtbl *ivtbl;
} u;
st_data_t index;
int iv_extended;
};
void
Init_var_tables(void)
{
rb_global_tbl = rb_id_table_create(0);
generic_iv_tbl_ = st_init_numtable();
autoload = rb_intern_const("__autoload__");
/* __classpath__: fully qualified class path */
classpath = rb_intern_const("__classpath__");
/* __tmp_classpath__: temporary class path which contains anonymous names */
tmp_classpath = rb_intern_const("__tmp_classpath__");
}
static inline bool
rb_namespace_p(VALUE obj)
{
if (RB_SPECIAL_CONST_P(obj)) return false;
switch (RB_BUILTIN_TYPE(obj)) {
case T_MODULE: case T_CLASS: return true;
2020-04-08 09:13:37 +03:00
default: break;
}
return false;
}
/**
* Returns +classpath+ of _klass_, if it is named, or +nil+ for
* anonymous +class+/+module+. A named +classpath+ may contain
* an anonymous component, but the last component is guaranteed
* to not be anonymous. <code>*permanent</code> is set to 1
* if +classpath+ has no anonymous components. There is no builtin
* Ruby level APIs that can change a permanent +classpath+.
*/
static VALUE
classname(VALUE klass, int *permanent)
{
st_table *ivtbl;
st_data_t n;
*permanent = 0;
if (!RCLASS_EXT(klass)) return Qnil;
if (!(ivtbl = RCLASS_IV_TBL(klass))) return Qnil;
if (st_lookup(ivtbl, (st_data_t)classpath, &n)) {
*permanent = 1;
return (VALUE)n;
}
if (st_lookup(ivtbl, (st_data_t)tmp_classpath, &n)) return (VALUE)n;
return Qnil;
}
/*
* call-seq:
* mod.name -> string
*
* Returns the name of the module <i>mod</i>. Returns nil for anonymous modules.
*/
VALUE
rb_mod_name(VALUE mod)
{
int permanent;
return classname(mod, &permanent);
}
static VALUE
make_temporary_path(VALUE obj, VALUE klass)
{
VALUE path;
switch (klass) {
case Qnil:
path = rb_sprintf("#<Class:%p>", (void*)obj);
break;
case Qfalse:
path = rb_sprintf("#<Module:%p>", (void*)obj);
break;
default:
path = rb_sprintf("#<%"PRIsVALUE":%p>", klass, (void*)obj);
break;
}
OBJ_FREEZE(path);
return path;
}
typedef VALUE (*fallback_func)(VALUE obj, VALUE name);
static VALUE
rb_tmp_class_path(VALUE klass, int *permanent, fallback_func fallback)
{
VALUE path = classname(klass, permanent);
if (!NIL_P(path)) {
return path;
}
else {
if (RB_TYPE_P(klass, T_MODULE)) {
if (rb_obj_class(klass) == rb_cModule) {
path = Qfalse;
}
else {
int perm;
path = rb_tmp_class_path(RBASIC(klass)->klass, &perm, fallback);
}
}
*permanent = 0;
return fallback(klass, path);
}
}
VALUE
rb_class_path(VALUE klass)
{
int permanent;
VALUE path = rb_tmp_class_path(klass, &permanent, make_temporary_path);
if (!NIL_P(path)) path = rb_str_dup(path);
return path;
}
VALUE
rb_class_path_cached(VALUE klass)
{
return rb_mod_name(klass);
}
static VALUE
no_fallback(VALUE obj, VALUE name)
{
return name;
}
VALUE
rb_search_class_path(VALUE klass)
{
int permanent;
return rb_tmp_class_path(klass, &permanent, no_fallback);
}
static VALUE
build_const_pathname(VALUE head, VALUE tail)
{
VALUE path = rb_str_dup(head);
rb_str_cat2(path, "::");
rb_str_append(path, tail);
return rb_fstring(path);
}
static VALUE
build_const_path(VALUE head, ID tail)
{
return build_const_pathname(head, rb_id2str(tail));
}
void
rb_set_class_path_string(VALUE klass, VALUE under, VALUE name)
{
VALUE str;
ID pathid = classpath;
if (under == rb_cObject) {
str = rb_str_new_frozen(name);
}
else {
int permanent;
str = rb_tmp_class_path(under, &permanent, make_temporary_path);
str = build_const_pathname(str, name);
if (!permanent) {
pathid = tmp_classpath;
}
}
rb_ivar_set(klass, pathid, str);
}
void
rb_set_class_path(VALUE klass, VALUE under, const char *name)
{
VALUE str = rb_str_new2(name);
OBJ_FREEZE(str);
rb_set_class_path_string(klass, under, str);
}
VALUE
rb_path_to_class(VALUE pathname)
{
rb_encoding *enc = rb_enc_get(pathname);
const char *pbeg, *pend, *p, *path = RSTRING_PTR(pathname);
ID id;
VALUE c = rb_cObject;
if (!rb_enc_asciicompat(enc)) {
rb_raise(rb_eArgError, "invalid class path encoding (non ASCII)");
}
pbeg = p = path;
pend = path + RSTRING_LEN(pathname);
if (path == pend || path[0] == '#') {
rb_raise(rb_eArgError, "can't retrieve anonymous class %"PRIsVALUE,
QUOTE(pathname));
}
while (p < pend) {
while (p < pend && *p != ':') p++;
id = rb_check_id_cstr(pbeg, p-pbeg, enc);
if (p < pend && p[0] == ':') {
if ((size_t)(pend - p) < 2 || p[1] != ':') goto undefined_class;
p += 2;
pbeg = p;
}
if (!id) {
goto undefined_class;
}
c = rb_const_search(c, id, TRUE, FALSE, FALSE);
if (c == Qundef) goto undefined_class;
if (!rb_namespace_p(c)) {
rb_raise(rb_eTypeError, "%"PRIsVALUE" does not refer to class/module",
pathname);
}
}
RB_GC_GUARD(pathname);
return c;
undefined_class:
rb_raise(rb_eArgError, "undefined class/module % "PRIsVALUE,
rb_str_subseq(pathname, 0, p-path));
UNREACHABLE_RETURN(Qundef);
}
VALUE
rb_path2class(const char *path)
{
return rb_path_to_class(rb_str_new_cstr(path));
}
VALUE
rb_class_name(VALUE klass)
{
return rb_class_path(rb_class_real(klass));
}
const char *
rb_class2name(VALUE klass)
{
int permanent;
VALUE path = rb_tmp_class_path(rb_class_real(klass), &permanent, make_temporary_path);
if (NIL_P(path)) return NULL;
return RSTRING_PTR(path);
}
const char *
rb_obj_classname(VALUE obj)
{
return rb_class2name(CLASS_OF(obj));
}
struct trace_var {
int removed;
void (*func)(VALUE arg, VALUE val);
VALUE data;
struct trace_var *next;
};
struct rb_global_variable {
int counter;
int block_trace;
VALUE *data;
rb_gvar_getter_t *getter;
rb_gvar_setter_t *setter;
rb_gvar_marker_t *marker;
rb_gvar_compact_t *compactor;
struct trace_var *trace;
};
struct rb_global_entry {
struct rb_global_variable *var;
ID id;
bool ractor_local;
};
static struct rb_global_entry*
rb_find_global_entry(ID id)
{
struct rb_global_entry *entry;
VALUE data;
if (!rb_id_table_lookup(rb_global_tbl, id, &data)) {
entry = NULL;
}
else {
entry = (struct rb_global_entry *)data;
RUBY_ASSERT(entry != NULL);
}
if (UNLIKELY(!rb_ractor_main_p()) && (!entry || !entry->ractor_local)) {
rb_raise(rb_eRactorIsolationError, "can not access global variables %s from non-main Ractors", rb_id2name(id));
}
return entry;
}
void
rb_gvar_ractor_local(const char *name)
{
struct rb_global_entry *entry = rb_find_global_entry(rb_intern(name));
entry->ractor_local = true;
}
static void
rb_gvar_undef_compactor(void *var)
{
}
static struct rb_global_entry*
rb_global_entry(ID id)
{
struct rb_global_entry *entry = rb_find_global_entry(id);
if (!entry) {
struct rb_global_variable *var;
entry = ALLOC(struct rb_global_entry);
var = ALLOC(struct rb_global_variable);
entry->id = id;
entry->var = var;
entry->ractor_local = false;
var->counter = 1;
var->data = 0;
var->getter = rb_gvar_undef_getter;
var->setter = rb_gvar_undef_setter;
var->marker = rb_gvar_undef_marker;
var->compactor = rb_gvar_undef_compactor;
var->block_trace = 0;
var->trace = 0;
rb_id_table_insert(rb_global_tbl, id, (VALUE)entry);
}
return entry;
}
VALUE
rb_gvar_undef_getter(ID id, VALUE *_)
{
rb_warning("global variable `%"PRIsVALUE"' not initialized", QUOTE_ID(id));
return Qnil;
}
static void
rb_gvar_val_compactor(void *_var)
{
struct rb_global_variable *var = (struct rb_global_variable *)_var;
VALUE obj = (VALUE)var->data;
if (obj) {
VALUE new = rb_gc_location(obj);
if (new != obj) {
var->data = (void*)new;
}
}
}
void
rb_gvar_undef_setter(VALUE val, ID id, VALUE *_)
{
struct rb_global_variable *var = rb_global_entry(id)->var;
var->getter = rb_gvar_val_getter;
var->setter = rb_gvar_val_setter;
var->marker = rb_gvar_val_marker;
var->compactor = rb_gvar_val_compactor;
var->data = (void*)val;
}
void
rb_gvar_undef_marker(VALUE *var)
{
}
VALUE
rb_gvar_val_getter(ID id, VALUE *data)
{
return (VALUE)data;
}
void
rb_gvar_val_setter(VALUE val, ID id, VALUE *_)
{
struct rb_global_variable *var = rb_global_entry(id)->var;
var->data = (void*)val;
}
void
rb_gvar_val_marker(VALUE *var)
{
VALUE data = (VALUE)var;
if (data) rb_gc_mark_movable(data);
}
VALUE
rb_gvar_var_getter(ID id, VALUE *var)
{
if (!var) return Qnil;
return *var;
}
void
rb_gvar_var_setter(VALUE val, ID id, VALUE *data)
{
*data = val;
}
void
rb_gvar_var_marker(VALUE *var)
{
if (var) rb_gc_mark_maybe(*var);
}
void
rb_gvar_readonly_setter(VALUE v, ID id, VALUE *_)
{
rb_name_error(id, "%"PRIsVALUE" is a read-only variable", QUOTE_ID(id));
}
static enum rb_id_table_iterator_result
mark_global_entry(VALUE v, void *ignored)
{
struct rb_global_entry *entry = (struct rb_global_entry *)v;
struct trace_var *trace;
struct rb_global_variable *var = entry->var;
(*var->marker)(var->data);
trace = var->trace;
while (trace) {
if (trace->data) rb_gc_mark_maybe(trace->data);
trace = trace->next;
}
return ID_TABLE_CONTINUE;
}
void
rb_gc_mark_global_tbl(void)
{
if (rb_global_tbl) {
rb_id_table_foreach_values(rb_global_tbl, mark_global_entry, 0);
}
}
static enum rb_id_table_iterator_result
update_global_entry(VALUE v, void *ignored)
{
struct rb_global_entry *entry = (struct rb_global_entry *)v;
struct rb_global_variable *var = entry->var;
(*var->compactor)(var);
return ID_TABLE_CONTINUE;
}
void
rb_gc_update_global_tbl(void)
{
if (rb_global_tbl) {
rb_id_table_foreach_values(rb_global_tbl, update_global_entry, 0);
}
}
static ID
global_id(const char *name)
{
ID id;
if (name[0] == '$') id = rb_intern(name);
else {
size_t len = strlen(name);
VALUE vbuf = 0;
char *buf = ALLOCV_N(char, vbuf, len+1);
buf[0] = '$';
memcpy(buf+1, name, len);
id = rb_intern2(buf, len+1);
ALLOCV_END(vbuf);
}
return id;
}
static ID
find_global_id(const char *name)
{
ID id;
size_t len = strlen(name);
if (name[0] == '$') {
id = rb_check_id_cstr(name, len, NULL);
}
else {
VALUE vbuf = 0;
char *buf = ALLOCV_N(char, vbuf, len+1);
buf[0] = '$';
memcpy(buf+1, name, len);
id = rb_check_id_cstr(buf, len+1, NULL);
ALLOCV_END(vbuf);
}
return id;
}
void
rb_define_hooked_variable(
const char *name,
VALUE *var,
rb_gvar_getter_t *getter,
rb_gvar_setter_t *setter)
{
volatile VALUE tmp = var ? *var : Qnil;
ID id = global_id(name);
struct rb_global_variable *gvar = rb_global_entry(id)->var;
gvar->data = (void*)var;
gvar->getter = getter ? (rb_gvar_getter_t *)getter : rb_gvar_var_getter;
gvar->setter = setter ? (rb_gvar_setter_t *)setter : rb_gvar_var_setter;
gvar->marker = rb_gvar_var_marker;
RB_GC_GUARD(tmp);
}
void
rb_define_variable(const char *name, VALUE *var)
{
rb_define_hooked_variable(name, var, 0, 0);
}
void
rb_define_readonly_variable(const char *name, const VALUE *var)
{
rb_define_hooked_variable(name, (VALUE *)var, 0, rb_gvar_readonly_setter);
}
void
rb_define_virtual_variable(
const char *name,
rb_gvar_getter_t *getter,
rb_gvar_setter_t *setter)
{
if (!getter) getter = rb_gvar_val_getter;
if (!setter) setter = rb_gvar_readonly_setter;
rb_define_hooked_variable(name, 0, getter, setter);
}
static void
rb_trace_eval(VALUE cmd, VALUE val)
{
rb_eval_cmd_kw(cmd, rb_ary_new3(1, val), RB_NO_KEYWORDS);
}
VALUE
rb_f_trace_var(int argc, const VALUE *argv)
{
VALUE var, cmd;
struct rb_global_entry *entry;
struct trace_var *trace;
if (rb_scan_args(argc, argv, "11", &var, &cmd) == 1) {
cmd = rb_block_proc();
}
if (NIL_P(cmd)) {
return rb_f_untrace_var(argc, argv);
}
entry = rb_global_entry(rb_to_id(var));
trace = ALLOC(struct trace_var);
trace->next = entry->var->trace;
trace->func = rb_trace_eval;
trace->data = cmd;
trace->removed = 0;
entry->var->trace = trace;
return Qnil;
}
static void
remove_trace(struct rb_global_variable *var)
{
struct trace_var *trace = var->trace;
struct trace_var t;
struct trace_var *next;
t.next = trace;
trace = &t;
while (trace->next) {
next = trace->next;
if (next->removed) {
trace->next = next->next;
xfree(next);
}
else {
trace = next;
}
}
var->trace = t.next;
}
VALUE
rb_f_untrace_var(int argc, const VALUE *argv)
{
VALUE var, cmd;
ID id;
struct rb_global_entry *entry;
struct trace_var *trace;
rb_scan_args(argc, argv, "11", &var, &cmd);
id = rb_check_id(&var);
if (!id) {
rb_name_error_str(var, "undefined global variable %"PRIsVALUE"", QUOTE(var));
}
if ((entry = rb_find_global_entry(id)) == NULL) {
rb_name_error(id, "undefined global variable %"PRIsVALUE"", QUOTE_ID(id));
}
trace = entry->var->trace;
if (NIL_P(cmd)) {
VALUE ary = rb_ary_new();
while (trace) {
struct trace_var *next = trace->next;
rb_ary_push(ary, (VALUE)trace->data);
trace->removed = 1;
trace = next;
}
if (!entry->var->block_trace) remove_trace(entry->var);
return ary;
}
else {
while (trace) {
if (trace->data == cmd) {
trace->removed = 1;
if (!entry->var->block_trace) remove_trace(entry->var);
return rb_ary_new3(1, cmd);
}
trace = trace->next;
}
}
return Qnil;
}
struct trace_data {
struct trace_var *trace;
VALUE val;
};
static VALUE
trace_ev(VALUE v)
{
struct trace_data *data = (void *)v;
struct trace_var *trace = data->trace;
while (trace) {
(*trace->func)(trace->data, data->val);
trace = trace->next;
}
2012-04-14 04:36:26 +04:00
return Qnil;
}
static VALUE
trace_en(VALUE v)
{
struct rb_global_variable *var = (void *)v;
var->block_trace = 0;
remove_trace(var);
return Qnil; /* not reached */
}
static VALUE
rb_gvar_set_entry(struct rb_global_entry *entry, VALUE val)
{
struct trace_data trace;
struct rb_global_variable *var = entry->var;
(*var->setter)(val, entry->id, var->data);
if (var->trace && !var->block_trace) {
var->block_trace = 1;
trace.trace = var->trace;
trace.val = val;
rb_ensure(trace_ev, (VALUE)&trace, trace_en, (VALUE)var);
}
return val;
}
VALUE
rb_gvar_set(ID id, VALUE val)
{
struct rb_global_entry *entry;
entry = rb_global_entry(id);
return rb_gvar_set_entry(entry, val);
}
VALUE
rb_gv_set(const char *name, VALUE val)
{
return rb_gvar_set(global_id(name), val);
}
VALUE
rb_gvar_get(ID id)
{
struct rb_global_entry *entry = rb_global_entry(id);
struct rb_global_variable *var = entry->var;
return (*var->getter)(entry->id, var->data);
}
VALUE
rb_gv_get(const char *name)
{
ID id = find_global_id(name);
2020-07-20 07:31:41 +03:00
if (!id) {
rb_warning("global variable `%s' not initialized", name);
return Qnil;
}
return rb_gvar_get(id);
}
mjit_compile.c: merge initial JIT compiler which has been developed by Takashi Kokubun <takashikkbn@gmail> as YARV-MJIT. Many of its bugs are fixed by wanabe <s.wanabe@gmail.com>. This JIT compiler is designed to be a safe migration path to introduce JIT compiler to MRI. So this commit does not include any bytecode changes or dynamic instruction modifications, which are done in original MJIT. This commit even strips off some aggressive optimizations from YARV-MJIT, and thus it's slower than YARV-MJIT too. But it's still fairly faster than Ruby 2.5 in some benchmarks (attached below). Note that this JIT compiler passes `make test`, `make test-all`, `make test-spec` without JIT, and even with JIT. Not only it's perfectly safe with JIT disabled because it does not replace VM instructions unlike MJIT, but also with JIT enabled it stably runs Ruby applications including Rails applications. I'm expecting this version as just "initial" JIT compiler. I have many optimization ideas which are skipped for initial merging, and you may easily replace this JIT compiler with a faster one by just replacing mjit_compile.c. `mjit_compile` interface is designed for the purpose. common.mk: update dependencies for mjit_compile.c. internal.h: declare `rb_vm_insn_addr2insn` for MJIT. vm.c: exclude some definitions if `-DMJIT_HEADER` is provided to compiler. This avoids to include some functions which take a long time to compile, e.g. vm_exec_core. Some of the purpose is achieved in transform_mjit_header.rb (see `IGNORED_FUNCTIONS`) but others are manually resolved for now. Load mjit_helper.h for MJIT header. mjit_helper.h: New. This is a file used only by JIT-ed code. I'll refactor `mjit_call_cfunc` later. vm_eval.c: add some #ifdef switches to skip compiling some functions like Init_vm_eval. win32/mkexports.rb: export thread/ec functions, which are used by MJIT. include/ruby/defines.h: add MJIT_FUNC_EXPORTED macro alis to clarify that a function is exported only for MJIT. array.c: export a function used by MJIT. bignum.c: ditto. class.c: ditto. compile.c: ditto. error.c: ditto. gc.c: ditto. hash.c: ditto. iseq.c: ditto. numeric.c: ditto. object.c: ditto. proc.c: ditto. re.c: ditto. st.c: ditto. string.c: ditto. thread.c: ditto. variable.c: ditto. vm_backtrace.c: ditto. vm_insnhelper.c: ditto. vm_method.c: ditto. I would like to improve maintainability of function exports, but I believe this way is acceptable as initial merging if we clarify the new exports are for MJIT (so that we can use them as TODO list to fix) and add unit tests to detect unresolved symbols. I'll add unit tests of JIT compilations in succeeding commits. Author: Takashi Kokubun <takashikkbn@gmail.com> Contributor: wanabe <s.wanabe@gmail.com> Part of [Feature #14235] --- * Known issues * Code generated by gcc is faster than clang. The benchmark may be worse in macOS. Following benchmark result is provided by gcc w/ Linux. * Performance is decreased when Google Chrome is running * JIT can work on MinGW, but it doesn't improve performance at least in short running benchmark. * Currently it doesn't perform well with Rails. We'll try to fix this before release. --- * Benchmark reslts Benchmarked with: Intel 4.0GHz i7-4790K with 16GB memory under x86-64 Ubuntu 8 Cores - 2.0.0-p0: Ruby 2.0.0-p0 - r62186: Ruby trunk (early 2.6.0), before MJIT changes - JIT off: On this commit, but without `--jit` option - JIT on: On this commit, and with `--jit` option ** Optcarrot fps Benchmark: https://github.com/mame/optcarrot | |2.0.0-p0 |r62186 |JIT off |JIT on | |:--------|:--------|:--------|:--------|:--------| |fps |37.32 |51.46 |51.31 |58.88 | |vs 2.0.0 |1.00x |1.38x |1.37x |1.58x | ** MJIT benchmarks Benchmark: https://github.com/benchmark-driver/mjit-benchmarks (Original: https://github.com/vnmakarov/ruby/tree/rtl_mjit_branch/MJIT-benchmarks) | |2.0.0-p0 |r62186 |JIT off |JIT on | |:----------|:--------|:--------|:--------|:--------| |aread |1.00 |1.09 |1.07 |2.19 | |aref |1.00 |1.13 |1.11 |2.22 | |aset |1.00 |1.50 |1.45 |2.64 | |awrite |1.00 |1.17 |1.13 |2.20 | |call |1.00 |1.29 |1.26 |2.02 | |const2 |1.00 |1.10 |1.10 |2.19 | |const |1.00 |1.11 |1.10 |2.19 | |fannk |1.00 |1.04 |1.02 |1.00 | |fib |1.00 |1.32 |1.31 |1.84 | |ivread |1.00 |1.13 |1.12 |2.43 | |ivwrite |1.00 |1.23 |1.21 |2.40 | |mandelbrot |1.00 |1.13 |1.16 |1.28 | |meteor |1.00 |2.97 |2.92 |3.17 | |nbody |1.00 |1.17 |1.15 |1.49 | |nest-ntimes|1.00 |1.22 |1.20 |1.39 | |nest-while |1.00 |1.10 |1.10 |1.37 | |norm |1.00 |1.18 |1.16 |1.24 | |nsvb |1.00 |1.16 |1.16 |1.17 | |red-black |1.00 |1.02 |0.99 |1.12 | |sieve |1.00 |1.30 |1.28 |1.62 | |trees |1.00 |1.14 |1.13 |1.19 | |while |1.00 |1.12 |1.11 |2.41 | ** Discourse's script/bench.rb Benchmark: https://github.com/discourse/discourse/blob/v1.8.7/script/bench.rb NOTE: Rails performance was somehow a little degraded with JIT for now. We should fix this. (At least I know opt_aref is performing badly in JIT and I have an idea to fix it. Please wait for the fix.) *** JIT off Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 17 75: 18 90: 22 99: 29 home_admin: 50: 21 75: 21 90: 27 99: 40 topic_admin: 50: 17 75: 18 90: 22 99: 32 categories: 50: 35 75: 41 90: 43 99: 77 home: 50: 39 75: 46 90: 49 99: 95 topic: 50: 46 75: 52 90: 56 99: 101 *** JIT on Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 19 75: 21 90: 25 99: 33 home_admin: 50: 24 75: 26 90: 30 99: 35 topic_admin: 50: 19 75: 20 90: 25 99: 30 categories: 50: 40 75: 44 90: 48 99: 76 home: 50: 42 75: 48 90: 51 99: 89 topic: 50: 49 75: 55 90: 58 99: 99 git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62197 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-02-04 14:22:28 +03:00
MJIT_FUNC_EXPORTED VALUE
rb_gvar_defined(ID id)
{
struct rb_global_entry *entry = rb_global_entry(id);
2021-09-15 02:11:05 +03:00
return RBOOL(entry->var->getter != rb_gvar_undef_getter);
}
rb_gvar_getter_t *
rb_gvar_getter_function_of(ID id)
{
const struct rb_global_entry *entry = rb_global_entry(id);
return entry->var->getter;
}
rb_gvar_setter_t *
rb_gvar_setter_function_of(ID id)
{
const struct rb_global_entry *entry = rb_global_entry(id);
return entry->var->setter;
}
static enum rb_id_table_iterator_result
gvar_i(ID key, VALUE val, void *a)
{
VALUE ary = (VALUE)a;
rb_ary_push(ary, ID2SYM(key));
return ID_TABLE_CONTINUE;
}
VALUE
rb_f_global_variables(void)
{
VALUE ary = rb_ary_new();
VALUE sym, backref = rb_backref_get();
if (!rb_ractor_main_p()) {
rb_raise(rb_eRactorIsolationError, "can not access global variables from non-main Ractors");
}
rb_id_table_foreach(rb_global_tbl, gvar_i, (void *)ary);
if (!NIL_P(backref)) {
char buf[2];
int i, nmatch = rb_match_count(backref);
buf[0] = '$';
for (i = 1; i <= nmatch; ++i) {
if (!rb_match_nth_defined(i, backref)) continue;
if (i < 10) {
/* probably reused, make static ID */
buf[1] = (char)(i + '0');
sym = ID2SYM(rb_intern2(buf, 2));
}
else {
/* dynamic symbol */
sym = rb_str_intern(rb_sprintf("$%d", i));
}
rb_ary_push(ary, sym);
}
}
return ary;
}
void
rb_alias_variable(ID name1, ID name2)
{
struct rb_global_entry *entry1, *entry2;
VALUE data1;
struct rb_id_table *gtbl = rb_global_tbl;
if (!rb_ractor_main_p()) {
rb_raise(rb_eRactorIsolationError, "can not access global variables from non-main Ractors");
}
entry2 = rb_global_entry(name2);
if (!rb_id_table_lookup(gtbl, name1, &data1)) {
entry1 = ALLOC(struct rb_global_entry);
entry1->id = name1;
rb_id_table_insert(gtbl, name1, (VALUE)entry1);
}
else if ((entry1 = (struct rb_global_entry *)data1)->var != entry2->var) {
struct rb_global_variable *var = entry1->var;
if (var->block_trace) {
rb_raise(rb_eRuntimeError, "can't alias in tracer");
}
var->counter--;
if (var->counter == 0) {
struct trace_var *trace = var->trace;
while (trace) {
struct trace_var *next = trace->next;
xfree(trace);
trace = next;
}
xfree(var);
}
}
else {
return;
}
entry2->var->counter++;
entry1->var = entry2->var;
}
static bool
iv_index_tbl_lookup(struct st_table *tbl, ID id, uint32_t *indexp)
{
2021-06-23 04:50:09 +03:00
st_data_t ent_data;
int r;
if (tbl == NULL) return false;
RB_VM_LOCK_ENTER();
{
2021-06-23 04:50:09 +03:00
r = st_lookup(tbl, (st_data_t)id, &ent_data);
}
RB_VM_LOCK_LEAVE();
if (r) {
2021-06-23 04:50:09 +03:00
struct rb_iv_index_tbl_entry *ent = (void *)ent_data;
*indexp = ent->index;
return true;
}
else {
return false;
}
}
static void
IVAR_ACCESSOR_SHOULD_BE_MAIN_RACTOR(ID id)
{
if (UNLIKELY(!rb_ractor_main_p())) {
if (rb_is_instance_id(id)) { // check only normal ivars
rb_raise(rb_eRactorIsolationError, "can not set instance variables of classes/modules by non-main Ractors");
}
}
}
#define CVAR_ACCESSOR_SHOULD_BE_MAIN_RACTOR() \
if (UNLIKELY(!rb_ractor_main_p())) { \
rb_raise(rb_eRactorIsolationError, "can not access class variables from non-main Ractors"); \
}
static inline struct st_table *
generic_ivtbl(VALUE obj, ID id, bool force_check_ractor)
{
ASSERT_vm_locking();
if ((force_check_ractor || LIKELY(rb_is_instance_id(id)) /* not internal ID */ ) &&
!RB_OBJ_FROZEN_RAW(obj) &&
UNLIKELY(!rb_ractor_main_p()) &&
UNLIKELY(rb_ractor_shareable_p(obj))) {
rb_raise(rb_eRactorIsolationError, "can not access instance variables of shareable objects from non-main Ractors");
}
return generic_iv_tbl_;
}
static inline struct st_table *
generic_ivtbl_no_ractor_check(VALUE obj)
{
return generic_ivtbl(obj, 0, false);
}
static int
gen_ivtbl_get(VALUE obj, ID id, struct gen_ivtbl **ivtbl)
{
st_data_t data;
int r = 0;
RB_VM_LOCK_ENTER();
{
if (st_lookup(generic_ivtbl(obj, id, false), (st_data_t)obj, &data)) {
*ivtbl = (struct gen_ivtbl *)data;
r = 1;
}
}
RB_VM_LOCK_LEAVE();
return r;
}
MJIT_FUNC_EXPORTED int
rb_ivar_generic_ivtbl_lookup(VALUE obj, struct gen_ivtbl **ivtbl)
{
return gen_ivtbl_get(obj, 0, ivtbl);
}
MJIT_FUNC_EXPORTED VALUE
rb_ivar_generic_lookup_with_index(VALUE obj, ID id, uint32_t index)
{
struct gen_ivtbl *ivtbl;
if (gen_ivtbl_get(obj, id, &ivtbl)) {
if (LIKELY(index < ivtbl->numiv)) {
VALUE val = ivtbl->ivptr[index];
return val;
}
}
return Qundef;
}
static VALUE
generic_ivar_delete(VALUE obj, ID id, VALUE undef)
{
struct gen_ivtbl *ivtbl;
if (gen_ivtbl_get(obj, id, &ivtbl)) {
st_table *iv_index_tbl = RCLASS_IV_INDEX_TBL(rb_obj_class(obj));
uint32_t index;
if (iv_index_tbl && iv_index_tbl_lookup(iv_index_tbl, id, &index)) {
if (index < ivtbl->numiv) {
VALUE ret = ivtbl->ivptr[index];
ivtbl->ivptr[index] = Qundef;
return ret == Qundef ? undef : ret;
}
}
}
return undef;
}
static VALUE
generic_ivar_get(VALUE obj, ID id, VALUE undef)
{
struct gen_ivtbl *ivtbl;
if (gen_ivtbl_get(obj, id, &ivtbl)) {
st_table *iv_index_tbl = RCLASS_IV_INDEX_TBL(rb_obj_class(obj));
uint32_t index;
if (iv_index_tbl && iv_index_tbl_lookup(iv_index_tbl, id, &index)) {
if (index < ivtbl->numiv) {
VALUE ret = ivtbl->ivptr[index];
return ret == Qundef ? undef : ret;
}
}
}
return undef;
}
static size_t
gen_ivtbl_bytes(size_t n)
{
return offsetof(struct gen_ivtbl, ivptr) + n * sizeof(VALUE);
}
static struct gen_ivtbl *
gen_ivtbl_resize(struct gen_ivtbl *old, uint32_t n)
{
uint32_t len = old ? old->numiv : 0;
struct gen_ivtbl *ivtbl = xrealloc(old, gen_ivtbl_bytes(n));
ivtbl->numiv = n;
for (; len < n; len++) {
ivtbl->ivptr[len] = Qundef;
}
return ivtbl;
}
#if 0
static struct gen_ivtbl *
gen_ivtbl_dup(const struct gen_ivtbl *orig)
{
size_t s = gen_ivtbl_bytes(orig->numiv);
struct gen_ivtbl *ivtbl = xmalloc(s);
memcpy(ivtbl, orig, s);
return ivtbl;
}
#endif
static uint32_t
iv_index_tbl_newsize(struct ivar_update *ivup)
{
if (!ivup->iv_extended) {
return (uint32_t)ivup->u.iv_index_tbl->num_entries;
}
else {
uint32_t index = (uint32_t)ivup->index; /* should not overflow */
return (index+1) + (index+1)/4; /* (index+1)*1.25 */
}
}
static int
generic_ivar_update(st_data_t *k, st_data_t *v, st_data_t u, int existing)
{
ASSERT_vm_locking();
struct ivar_update *ivup = (struct ivar_update *)u;
struct gen_ivtbl *ivtbl = 0;
if (existing) {
ivtbl = (struct gen_ivtbl *)*v;
if (ivup->index < ivtbl->numiv) {
ivup->u.ivtbl = ivtbl;
return ST_STOP;
}
}
FL_SET((VALUE)*k, FL_EXIVAR);
uint32_t newsize = iv_index_tbl_newsize(ivup);
ivtbl = gen_ivtbl_resize(ivtbl, newsize);
*v = (st_data_t)ivtbl;
ivup->u.ivtbl = ivtbl;
return ST_CONTINUE;
}
static VALUE
generic_ivar_defined(VALUE obj, ID id)
{
struct gen_ivtbl *ivtbl;
st_table *iv_index_tbl = RCLASS_IV_INDEX_TBL(rb_obj_class(obj));
uint32_t index;
if (!iv_index_tbl_lookup(iv_index_tbl, id, &index)) return Qfalse;
if (!gen_ivtbl_get(obj, id, &ivtbl)) return Qfalse;
2021-09-15 02:11:05 +03:00
return RBOOL((index < ivtbl->numiv) && (ivtbl->ivptr[index] != Qundef));
}
static int
generic_ivar_remove(VALUE obj, ID id, VALUE *valp)
{
struct gen_ivtbl *ivtbl;
uint32_t index;
st_table *iv_index_tbl = RCLASS_IV_INDEX_TBL(rb_obj_class(obj));
if (!iv_index_tbl) return 0;
if (!iv_index_tbl_lookup(iv_index_tbl, id, &index)) return 0;
if (!gen_ivtbl_get(obj, id, &ivtbl)) return 0;
if (index < ivtbl->numiv) {
if (ivtbl->ivptr[index] != Qundef) {
*valp = ivtbl->ivptr[index];
ivtbl->ivptr[index] = Qundef;
return 1;
}
}
return 0;
}
static void
gen_ivtbl_mark(const struct gen_ivtbl *ivtbl)
{
uint32_t i;
for (i = 0; i < ivtbl->numiv; i++) {
rb_gc_mark(ivtbl->ivptr[i]);
}
}
void
rb_mark_generic_ivar(VALUE obj)
{
struct gen_ivtbl *ivtbl;
if (gen_ivtbl_get(obj, 0, &ivtbl)) {
gen_ivtbl_mark(ivtbl);
}
}
void
rb_mv_generic_ivar(VALUE rsrc, VALUE dst)
{
st_data_t key = (st_data_t)rsrc;
2021-06-23 04:50:09 +03:00
st_data_t ivtbl;
2021-06-23 04:50:09 +03:00
if (st_delete(generic_ivtbl_no_ractor_check(rsrc), &key, &ivtbl))
st_insert(generic_ivtbl_no_ractor_check(dst), (st_data_t)dst, ivtbl);
}
void
rb_free_generic_ivar(VALUE obj)
{
2021-06-23 04:50:09 +03:00
st_data_t key = (st_data_t)obj, ivtbl;
2021-06-23 04:50:09 +03:00
if (st_delete(generic_ivtbl_no_ractor_check(obj), &key, &ivtbl))
xfree((struct gen_ivtbl *)ivtbl);
}
RUBY_FUNC_EXPORTED size_t
rb_generic_ivar_memsize(VALUE obj)
{
struct gen_ivtbl *ivtbl;
if (gen_ivtbl_get(obj, 0, &ivtbl))
return gen_ivtbl_bytes(ivtbl->numiv);
return 0;
}
static size_t
gen_ivtbl_count(const struct gen_ivtbl *ivtbl)
{
uint32_t i;
size_t n = 0;
for (i = 0; i < ivtbl->numiv; i++) {
if (ivtbl->ivptr[i] != Qundef) {
n++;
}
}
return n;
}
static int
lock_st_lookup(st_table *tab, st_data_t key, st_data_t *value)
{
int r;
RB_VM_LOCK_ENTER();
{
r = st_lookup(tab, key, value);
}
RB_VM_LOCK_LEAVE();
return r;
}
static int
lock_st_delete(st_table *tab, st_data_t *key, st_data_t *value)
{
int r;
RB_VM_LOCK_ENTER();
{
r = st_delete(tab, key, value);
}
RB_VM_LOCK_LEAVE();
return r;
}
static int
lock_st_is_member(st_table *tab, st_data_t key)
{
int r;
RB_VM_LOCK_ENTER();
{
r = st_is_member(tab, key);
}
RB_VM_LOCK_LEAVE();
return r;
}
static int
lock_st_insert(st_table *tab, st_data_t key, st_data_t value)
{
int r;
RB_VM_LOCK_ENTER();
{
r = st_insert(tab, key, value);
}
RB_VM_LOCK_LEAVE();
return r;
}
VALUE
rb_ivar_lookup(VALUE obj, ID id, VALUE undef)
{
if (SPECIAL_CONST_P(obj)) return undef;
switch (BUILTIN_TYPE(obj)) {
case T_OBJECT:
{
uint32_t index;
uint32_t len = ROBJECT_NUMIV(obj);
VALUE *ptr = ROBJECT_IVPTR(obj);
2021-06-23 04:50:09 +03:00
VALUE val;
if (iv_index_tbl_lookup(ROBJECT_IV_INDEX_TBL(obj), id, &index) &&
index < len &&
(val = ptr[index]) != Qundef) {
return val;
}
else {
break;
}
}
case T_CLASS:
case T_MODULE:
{
2021-06-23 04:50:09 +03:00
st_data_t val;
if (RCLASS_IV_TBL(obj) &&
lock_st_lookup(RCLASS_IV_TBL(obj), (st_data_t)id, &val)) {
if (rb_is_instance_id(id) &&
UNLIKELY(!rb_ractor_main_p()) &&
!rb_ractor_shareable_p(val)) {
rb_raise(rb_eRactorIsolationError,
"can not get unshareable values from instance variables of classes/modules from non-main Ractors");
}
return val;
}
else {
break;
}
}
default:
if (FL_TEST(obj, FL_EXIVAR))
return generic_ivar_get(obj, id, undef);
break;
}
return undef;
}
VALUE
rb_ivar_get(VALUE obj, ID id)
{
VALUE iv = rb_ivar_lookup(obj, id, Qnil);
RB_DEBUG_COUNTER_INC(ivar_get_base);
return iv;
}
VALUE
rb_attr_get(VALUE obj, ID id)
{
return rb_ivar_lookup(obj, id, Qnil);
}
static VALUE
rb_ivar_delete(VALUE obj, ID id, VALUE undef)
{
2021-06-23 04:50:09 +03:00
VALUE *ptr;
struct st_table *iv_index_tbl;
uint32_t len, index;
rb_check_frozen(obj);
switch (BUILTIN_TYPE(obj)) {
case T_OBJECT:
len = ROBJECT_NUMIV(obj);
ptr = ROBJECT_IVPTR(obj);
iv_index_tbl = ROBJECT_IV_INDEX_TBL(obj);
if (iv_index_tbl_lookup(iv_index_tbl, id, &index) &&
index < len) {
2021-06-23 04:50:09 +03:00
VALUE val = ptr[index];
ptr[index] = Qundef;
if (val != Qundef) {
return val;
}
}
break;
case T_CLASS:
case T_MODULE:
IVAR_ACCESSOR_SHOULD_BE_MAIN_RACTOR(id);
2021-06-23 04:50:09 +03:00
if (RCLASS_IV_TBL(obj)) {
st_data_t id_data = (st_data_t)id, val;
if (lock_st_delete(RCLASS_IV_TBL(obj), &id_data, &val)) {
2021-06-23 04:50:09 +03:00
return (VALUE)val;
}
}
break;
default:
if (FL_TEST(obj, FL_EXIVAR))
return generic_ivar_delete(obj, id, undef);
break;
}
return undef;
}
VALUE
rb_attr_delete(VALUE obj, ID id)
{
return rb_ivar_delete(obj, id, Qnil);
}
static st_table *
iv_index_tbl_make(VALUE obj, VALUE klass)
{
st_table *iv_index_tbl;
if (UNLIKELY(!klass)) {
rb_raise(rb_eTypeError, "hidden object cannot have instance variables");
}
if ((iv_index_tbl = RCLASS_IV_INDEX_TBL(klass)) == NULL) {
RB_VM_LOCK_ENTER();
if ((iv_index_tbl = RCLASS_IV_INDEX_TBL(klass)) == NULL) {
iv_index_tbl = RCLASS_IV_INDEX_TBL(klass) = st_init_numtable();
}
RB_VM_LOCK_LEAVE();
}
return iv_index_tbl;
}
static void
iv_index_tbl_extend(struct ivar_update *ivup, ID id, VALUE klass)
{
ASSERT_vm_locking();
2021-06-23 04:50:09 +03:00
st_data_t ent_data;
struct rb_iv_index_tbl_entry *ent;
2021-06-23 04:50:09 +03:00
if (st_lookup(ivup->u.iv_index_tbl, (st_data_t)id, &ent_data)) {
ent = (void *)ent_data;
ivup->index = ent->index;
return;
}
if (ivup->u.iv_index_tbl->num_entries >= INT_MAX) {
rb_raise(rb_eArgError, "too many instance variables");
}
ent = ALLOC(struct rb_iv_index_tbl_entry);
ent->index = ivup->index = (uint32_t)ivup->u.iv_index_tbl->num_entries;
ent->class_value = klass;
ent->class_serial = RCLASS_SERIAL(klass);
st_add_direct(ivup->u.iv_index_tbl, (st_data_t)id, (st_data_t)ent);
ivup->iv_extended = 1;
}
static void
generic_ivar_set(VALUE obj, ID id, VALUE val)
{
VALUE klass = rb_obj_class(obj);
struct ivar_update ivup;
ivup.iv_extended = 0;
ivup.u.iv_index_tbl = iv_index_tbl_make(obj, klass);
RB_VM_LOCK_ENTER();
{
iv_index_tbl_extend(&ivup, id, klass);
st_update(generic_ivtbl(obj, id, false), (st_data_t)obj, generic_ivar_update,
(st_data_t)&ivup);
}
RB_VM_LOCK_LEAVE();
ivup.u.ivtbl->ivptr[ivup.index] = val;
RB_OBJ_WRITTEN(obj, Qundef, val);
}
static VALUE *
obj_ivar_heap_alloc(VALUE obj, size_t newsize)
{
VALUE *newptr = rb_transient_heap_alloc(obj, sizeof(VALUE) * newsize);
if (newptr != NULL) {
ROBJ_TRANSIENT_SET(obj);
}
else {
ROBJ_TRANSIENT_UNSET(obj);
newptr = ALLOC_N(VALUE, newsize);
}
return newptr;
}
static VALUE *
obj_ivar_heap_realloc(VALUE obj, int32_t len, size_t newsize)
{
VALUE *newptr;
int i;
if (ROBJ_TRANSIENT_P(obj)) {
const VALUE *orig_ptr = ROBJECT(obj)->as.heap.ivptr;
newptr = obj_ivar_heap_alloc(obj, newsize);
assert(newptr);
ROBJECT(obj)->as.heap.ivptr = newptr;
for (i=0; i<(int)len; i++) {
newptr[i] = orig_ptr[i];
}
}
else {
REALLOC_N(ROBJECT(obj)->as.heap.ivptr, VALUE, newsize);
newptr = ROBJECT(obj)->as.heap.ivptr;
}
return newptr;
}
#if USE_TRANSIENT_HEAP
void
rb_obj_transient_heap_evacuate(VALUE obj, int promote)
{
if (ROBJ_TRANSIENT_P(obj)) {
uint32_t len = ROBJECT_NUMIV(obj);
const VALUE *old_ptr = ROBJECT_IVPTR(obj);
VALUE *new_ptr;
if (promote) {
new_ptr = ALLOC_N(VALUE, len);
ROBJ_TRANSIENT_UNSET(obj);
}
else {
new_ptr = obj_ivar_heap_alloc(obj, len);
}
MEMCPY(new_ptr, old_ptr, VALUE, len);
ROBJECT(obj)->as.heap.ivptr = new_ptr;
}
}
#endif
static void
init_iv_list(VALUE obj, uint32_t len, uint32_t newsize, st_table *index_tbl)
{
VALUE *ptr = ROBJECT_IVPTR(obj);
VALUE *newptr;
if (RBASIC(obj)->flags & ROBJECT_EMBED) {
newptr = obj_ivar_heap_alloc(obj, newsize);
MEMCPY(newptr, ptr, VALUE, len);
RBASIC(obj)->flags &= ~ROBJECT_EMBED;
ROBJECT(obj)->as.heap.ivptr = newptr;
}
else {
newptr = obj_ivar_heap_realloc(obj, len, newsize);
}
for (; len < newsize; len++) {
newptr[len] = Qundef;
}
ROBJECT(obj)->as.heap.numiv = newsize;
ROBJECT(obj)->as.heap.iv_index_tbl = index_tbl;
}
void
rb_init_iv_list(VALUE obj)
{
st_table *index_tbl = ROBJECT_IV_INDEX_TBL(obj);
uint32_t newsize = (uint32_t)index_tbl->num_entries;
uint32_t len = ROBJECT_NUMIV(obj);
init_iv_list(obj, len, newsize, index_tbl);
}
// Retreive or create the id-to-index mapping for a given object and an
// instance variable name.
static struct ivar_update
obj_ensure_iv_index_mapping(VALUE obj, ID id)
{
VALUE klass = rb_obj_class(obj);
struct ivar_update ivup;
ivup.iv_extended = 0;
ivup.u.iv_index_tbl = iv_index_tbl_make(obj, klass);
RB_VM_LOCK_ENTER();
{
iv_index_tbl_extend(&ivup, id, klass);
}
RB_VM_LOCK_LEAVE();
return ivup;
}
// Return the instance variable index for a given name and T_OBJECT object. The
// mapping between name and index lives on `rb_obj_class(obj)` and is created
// if not already present.
//
// @note May raise when there are too many instance variables.
// @note YJIT uses this function at compile time to simplify the work needed to
// access the variable at runtime.
uint32_t
rb_obj_ensure_iv_index_mapping(VALUE obj, ID id)
{
RUBY_ASSERT(RB_TYPE_P(obj, T_OBJECT));
// This uint32_t cast shouldn't lose information as it's checked in
// iv_index_tbl_extend(). The index is stored as an uint32_t in
// struct rb_iv_index_tbl_entry.
return (uint32_t)obj_ensure_iv_index_mapping(obj, id).index;
}
static VALUE
obj_ivar_set(VALUE obj, ID id, VALUE val)
{
uint32_t len;
struct ivar_update ivup = obj_ensure_iv_index_mapping(obj, id);
len = ROBJECT_NUMIV(obj);
if (len <= ivup.index) {
uint32_t newsize = iv_index_tbl_newsize(&ivup);
init_iv_list(obj, len, newsize, ivup.u.iv_index_tbl);
}
RB_OBJ_WRITE(obj, &ROBJECT_IVPTR(obj)[ivup.index], val);
return val;
}
static void
ivar_set(VALUE obj, ID id, VALUE val)
{
RB_DEBUG_COUNTER_INC(ivar_set_base);
switch (BUILTIN_TYPE(obj)) {
case T_OBJECT:
obj_ivar_set(obj, id, val);
break;
case T_CLASS:
case T_MODULE:
IVAR_ACCESSOR_SHOULD_BE_MAIN_RACTOR(id);
rb_class_ivar_set(obj, id, val);
break;
default:
generic_ivar_set(obj, id, val);
break;
}
}
VALUE
rb_ivar_set(VALUE obj, ID id, VALUE val)
{
rb_check_frozen(obj);
ivar_set(obj, id, val);
return val;
}
void
rb_ivar_set_internal(VALUE obj, ID id, VALUE val)
{
// should be internal instance variable name (no @ prefix)
VM_ASSERT(!rb_is_instance_id(id));
ivar_set(obj, id, val);
}
VALUE
rb_ivar_defined(VALUE obj, ID id)
{
VALUE val;
struct st_table *iv_index_tbl;
uint32_t index;
if (SPECIAL_CONST_P(obj)) return Qfalse;
switch (BUILTIN_TYPE(obj)) {
case T_OBJECT:
iv_index_tbl = ROBJECT_IV_INDEX_TBL(obj);
if (iv_index_tbl_lookup(iv_index_tbl, id, &index) &&
index < ROBJECT_NUMIV(obj) &&
(val = ROBJECT_IVPTR(obj)[index]) != Qundef) {
return Qtrue;
}
break;
case T_CLASS:
case T_MODULE:
if (RCLASS_IV_TBL(obj) && lock_st_is_member(RCLASS_IV_TBL(obj), (st_data_t)id))
return Qtrue;
break;
default:
if (FL_TEST(obj, FL_EXIVAR))
return generic_ivar_defined(obj, id);
break;
}
return Qfalse;
}
typedef int rb_ivar_foreach_callback_func(ID key, VALUE val, st_data_t arg);
st_data_t rb_st_nth_key(st_table *tab, st_index_t index);
static ID
iv_index_tbl_nth_id(st_table *iv_index_tbl, uint32_t index)
{
st_data_t key;
RB_VM_LOCK_ENTER();
{
key = rb_st_nth_key(iv_index_tbl, index);
}
RB_VM_LOCK_LEAVE();
return (ID)key;
}
static inline bool
ivar_each_i(st_table *iv_index_tbl, VALUE val, uint32_t i, rb_ivar_foreach_callback_func *func, st_data_t arg)
{
if (val != Qundef) {
ID id = iv_index_tbl_nth_id(iv_index_tbl, i);
switch (func(id, val, arg)) {
case ST_CHECK:
case ST_CONTINUE:
break;
case ST_STOP:
return true;
default:
rb_bug("unreachable");
}
}
return false;
}
static void
obj_ivar_each(VALUE obj, rb_ivar_foreach_callback_func *func, st_data_t arg)
{
st_table *iv_index_tbl = ROBJECT_IV_INDEX_TBL(obj);
if (!iv_index_tbl) return;
uint32_t i=0;
for (i=0; i < ROBJECT_NUMIV(obj); i++) {
VALUE val = ROBJECT_IVPTR(obj)[i];
if (ivar_each_i(iv_index_tbl, val, i, func, arg)) {
return;
}
}
}
static void
gen_ivar_each(VALUE obj, rb_ivar_foreach_callback_func *func, st_data_t arg)
{
struct gen_ivtbl *ivtbl;
st_table *iv_index_tbl = RCLASS_IV_INDEX_TBL(rb_obj_class(obj));
if (!iv_index_tbl) return;
if (!gen_ivtbl_get(obj, 0, &ivtbl)) return;
for (uint32_t i=0; i<ivtbl->numiv; i++) {
VALUE val = ivtbl->ivptr[i];
if (ivar_each_i(iv_index_tbl, val, i, func, arg)) {
return;
}
}
}
struct givar_copy {
VALUE obj;
VALUE klass;
st_table *iv_index_tbl;
struct gen_ivtbl *ivtbl;
};
static int
gen_ivar_copy(ID id, VALUE val, st_data_t arg)
{
struct givar_copy *c = (struct givar_copy *)arg;
struct ivar_update ivup;
ivup.iv_extended = 0;
ivup.u.iv_index_tbl = c->iv_index_tbl;
RB_VM_LOCK_ENTER();
{
iv_index_tbl_extend(&ivup, id, c->klass);
}
RB_VM_LOCK_LEAVE();
if (ivup.index >= c->ivtbl->numiv) {
uint32_t newsize = iv_index_tbl_newsize(&ivup);
c->ivtbl = gen_ivtbl_resize(c->ivtbl, newsize);
}
c->ivtbl->ivptr[ivup.index] = val;
RB_OBJ_WRITTEN(c->obj, Qundef, val);
return ST_CONTINUE;
}
void
rb_copy_generic_ivar(VALUE clone, VALUE obj)
{
struct gen_ivtbl *ivtbl;
rb_check_frozen(clone);
if (!FL_TEST(obj, FL_EXIVAR)) {
goto clear;
}
if (gen_ivtbl_get(obj, 0, &ivtbl)) {
struct givar_copy c;
uint32_t i;
if (gen_ivtbl_count(ivtbl) == 0)
goto clear;
if (gen_ivtbl_get(clone, 0, &c.ivtbl)) {
for (i = 0; i < c.ivtbl->numiv; i++)
c.ivtbl->ivptr[i] = Qundef;
}
else {
c.ivtbl = gen_ivtbl_resize(0, ivtbl->numiv);
FL_SET(clone, FL_EXIVAR);
}
VALUE klass = rb_obj_class(clone);
c.iv_index_tbl = iv_index_tbl_make(clone, klass);
c.obj = clone;
c.klass = klass;
gen_ivar_each(obj, gen_ivar_copy, (st_data_t)&c);
/*
* c.ivtbl may change in gen_ivar_copy due to realloc,
* no need to free
*/
RB_VM_LOCK_ENTER();
{
generic_ivtbl_no_ractor_check(clone);
st_insert(generic_ivtbl_no_ractor_check(obj), (st_data_t)clone, (st_data_t)c.ivtbl);
}
RB_VM_LOCK_LEAVE();
}
return;
clear:
if (FL_TEST(clone, FL_EXIVAR)) {
rb_free_generic_ivar(clone);
FL_UNSET(clone, FL_EXIVAR);
}
}
void
rb_replace_generic_ivar(VALUE clone, VALUE obj)
{
RUBY_ASSERT(FL_TEST(obj, FL_EXIVAR));
RB_VM_LOCK_ENTER();
{
2021-06-23 04:50:09 +03:00
st_data_t ivtbl, obj_data = (st_data_t)obj;
if (st_lookup(generic_iv_tbl_, (st_data_t)obj, &ivtbl)) {
st_insert(generic_iv_tbl_, (st_data_t)clone, ivtbl);
st_delete(generic_iv_tbl_, &obj_data, NULL);
}
else {
rb_bug("unreachable");
}
}
RB_VM_LOCK_LEAVE();
FL_SET(clone, FL_EXIVAR);
}
void
rb_ivar_foreach(VALUE obj, rb_ivar_foreach_callback_func *func, st_data_t arg)
{
if (SPECIAL_CONST_P(obj)) return;
switch (BUILTIN_TYPE(obj)) {
case T_OBJECT:
obj_ivar_each(obj, func, arg);
break;
case T_CLASS:
case T_MODULE:
IVAR_ACCESSOR_SHOULD_BE_MAIN_RACTOR(0);
if (RCLASS_IV_TBL(obj)) {
RB_VM_LOCK_ENTER();
{
st_foreach_safe(RCLASS_IV_TBL(obj), func, arg);
}
RB_VM_LOCK_LEAVE();
}
break;
default:
if (FL_TEST(obj, FL_EXIVAR)) {
gen_ivar_each(obj, func, arg);
}
break;
}
}
st_index_t
rb_ivar_count(VALUE obj)
{
st_table *tbl;
if (SPECIAL_CONST_P(obj)) return 0;
switch (BUILTIN_TYPE(obj)) {
case T_OBJECT:
if (ROBJECT_IV_INDEX_TBL(obj) != 0) {
st_index_t i, count, num = ROBJECT_NUMIV(obj);
const VALUE *const ivptr = ROBJECT_IVPTR(obj);
for (i = count = 0; i < num; ++i) {
if (ivptr[i] != Qundef) {
count++;
}
}
return count;
}
break;
case T_CLASS:
case T_MODULE:
if ((tbl = RCLASS_IV_TBL(obj)) != 0) {
return tbl->num_entries;
}
break;
default:
if (FL_TEST(obj, FL_EXIVAR)) {
struct gen_ivtbl *ivtbl;
if (gen_ivtbl_get(obj, 0, &ivtbl)) {
return gen_ivtbl_count(ivtbl);
}
}
break;
}
return 0;
}
static int
ivar_i(st_data_t k, st_data_t v, st_data_t a)
{
ID key = (ID)k;
VALUE ary = (VALUE)a;
if (rb_is_instance_id(key)) {
rb_ary_push(ary, ID2SYM(key));
}
return ST_CONTINUE;
}
/*
* call-seq:
* obj.instance_variables -> array
*
* Returns an array of instance variable names for the receiver. Note
* that simply defining an accessor does not create the corresponding
* instance variable.
*
* class Fred
* attr_accessor :a1
* def initialize
* @iv = 3
* end
* end
* Fred.new.instance_variables #=> [:@iv]
*/
VALUE
rb_obj_instance_variables(VALUE obj)
{
VALUE ary;
ary = rb_ary_new();
rb_ivar_foreach(obj, ivar_i, ary);
return ary;
}
#define rb_is_constant_id rb_is_const_id
#define rb_is_constant_name rb_is_const_name
#define id_for_var(obj, name, part, type) \
id_for_var_message(obj, name, type, "`%1$s' is not allowed as "#part" "#type" variable name")
#define id_for_var_message(obj, name, type, message) \
check_id_type(obj, &(name), rb_is_##type##_id, rb_is_##type##_name, message, strlen(message))
static ID
check_id_type(VALUE obj, VALUE *pname,
int (*valid_id_p)(ID), int (*valid_name_p)(VALUE),
const char *message, size_t message_len)
{
ID id = rb_check_id(pname);
VALUE name = *pname;
if (id ? !valid_id_p(id) : !valid_name_p(name)) {
rb_name_err_raise_str(rb_fstring_new(message, message_len),
obj, name);
}
return id;
}
/*
* call-seq:
* obj.remove_instance_variable(symbol) -> obj
* obj.remove_instance_variable(string) -> obj
*
* Removes the named instance variable from <i>obj</i>, returning that
* variable's value.
* String arguments are converted to symbols.
*
* class Dummy
* attr_reader :var
* def initialize
* @var = 99
* end
* def remove
* remove_instance_variable(:@var)
* end
* end
* d = Dummy.new
* d.var #=> 99
* d.remove #=> 99
* d.var #=> nil
*/
VALUE
rb_obj_remove_instance_variable(VALUE obj, VALUE name)
{
VALUE val = Qnil;
const ID id = id_for_var(obj, name, an, instance);
st_data_t n, v;
struct st_table *iv_index_tbl;
uint32_t index;
rb_check_frozen(obj);
if (!id) {
goto not_defined;
}
switch (BUILTIN_TYPE(obj)) {
case T_OBJECT:
iv_index_tbl = ROBJECT_IV_INDEX_TBL(obj);
if (iv_index_tbl_lookup(iv_index_tbl, id, &index) &&
index < ROBJECT_NUMIV(obj) &&
(val = ROBJECT_IVPTR(obj)[index]) != Qundef) {
ROBJECT_IVPTR(obj)[index] = Qundef;
return val;
}
break;
case T_CLASS:
case T_MODULE:
IVAR_ACCESSOR_SHOULD_BE_MAIN_RACTOR(id);
n = id;
if (RCLASS_IV_TBL(obj) && lock_st_delete(RCLASS_IV_TBL(obj), &n, &v)) {
return (VALUE)v;
}
break;
default:
if (FL_TEST(obj, FL_EXIVAR)) {
if (generic_ivar_remove(obj, id, &val)) {
return val;
}
}
break;
}
2012-04-14 04:36:26 +04:00
not_defined:
rb_name_err_raise("instance variable %1$s not defined",
obj, name);
UNREACHABLE_RETURN(Qnil);
}
NORETURN(static void uninitialized_constant(VALUE, VALUE));
static void
uninitialized_constant(VALUE klass, VALUE name)
{
if (klass && rb_class_real(klass) != rb_cObject)
rb_name_err_raise("uninitialized constant %2$s::%1$s",
klass, name);
else
rb_name_err_raise("uninitialized constant %1$s",
klass, name);
}
VALUE
rb_const_missing(VALUE klass, VALUE name)
{
VALUE value = rb_funcallv(klass, idConst_missing, 1, &name);
rb_vm_inc_const_missing_count();
return value;
}
/*
* call-seq:
* mod.const_missing(sym) -> obj
*
* Invoked when a reference is made to an undefined constant in
* <i>mod</i>. It is passed a symbol for the undefined constant, and
* returns a value to be used for that constant. The
* following code is an example of the same:
*
* def Foo.const_missing(name)
* name # return the constant name as Symbol
* end
*
* Foo::UNDEFINED_CONST #=> :UNDEFINED_CONST: symbol returned
*
* In the next example when a reference is made to an undefined constant,
* it attempts to load a file whose name is the lowercase version of the
* constant (thus class <code>Fred</code> is assumed to be in file
* <code>fred.rb</code>). If found, it returns the loaded class. It
* therefore implements an autoload feature similar to Kernel#autoload and
* Module#autoload.
*
* def Object.const_missing(name)
* @looked_for ||= {}
* str_name = name.to_s
* raise "Class not found: #{name}" if @looked_for[str_name]
* @looked_for[str_name] = 1
* file = str_name.downcase
* require file
* klass = const_get(name)
* return klass if klass
* raise "Class not found: #{name}"
* end
*
*/
VALUE
rb_mod_const_missing(VALUE klass, VALUE name)
{
VALUE ref = GET_EC()->private_const_reference;
rb_vm_pop_cfunc_frame();
if (ref) {
rb_name_err_raise("private constant %2$s::%1$s referenced",
ref, name);
}
uninitialized_constant(klass, name);
2012-04-14 04:36:26 +04:00
UNREACHABLE_RETURN(Qnil);
}
static void
autoload_mark(void *ptr)
{
rb_mark_tbl_no_pin((st_table *)ptr);
}
static void
autoload_free(void *ptr)
{
st_free_table((st_table *)ptr);
}
static size_t
autoload_memsize(const void *ptr)
{
const st_table *tbl = ptr;
return st_memsize(tbl);
}
static void
autoload_compact(void *ptr)
{
rb_gc_update_tbl_refs((st_table *)ptr);
}
static const rb_data_type_t autoload_data_type = {
"autoload",
{autoload_mark, autoload_free, autoload_memsize, autoload_compact,},
0, 0, RUBY_TYPED_FREE_IMMEDIATELY
};
#define check_autoload_table(av) \
(struct st_table *)rb_check_typeddata((av), &autoload_data_type)
static VALUE
autoload_data(VALUE mod, ID id)
{
struct st_table *tbl;
st_data_t val;
if (!st_lookup(RCLASS_IV_TBL(mod), autoload, &val) ||
!(tbl = check_autoload_table((VALUE)val)) ||
!st_lookup(tbl, (st_data_t)id, &val)) {
return 0;
}
return (VALUE)val;
}
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
struct autoload_const {
struct list_node cnode; /* <=> autoload_data_i.constants */
VALUE mod;
VALUE ad; /* autoload_data_i */
VALUE value;
VALUE file;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
ID id;
rb_const_flag_t flag;
int line;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
};
/* always on stack, no need to mark */
struct autoload_state {
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
struct autoload_const *ac;
VALUE result;
VALUE thread;
2021-06-23 04:50:09 +03:00
struct list_head waitq;
};
struct autoload_data_i {
VALUE feature;
struct autoload_state *state; /* points to on-stack struct */
rb_serial_t fork_gen;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
struct list_head constants; /* <=> autoload_const.cnode */
};
static void
autoload_i_compact(void *ptr)
{
struct autoload_data_i *p = ptr;
p->feature = rb_gc_location(p->feature);
}
static void
autoload_i_mark(void *ptr)
{
struct autoload_data_i *p = ptr;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
rb_gc_mark_movable(p->feature);
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
/* allow GC to free us if no modules refer to this via autoload_const.ad */
if (list_empty(&p->constants)) {
rb_hash_delete(autoload_featuremap, p->feature);
}
}
static void
autoload_i_free(void *ptr)
{
struct autoload_data_i *p = ptr;
/* we may leak some memory at VM shutdown time, no big deal */
if (list_empty(&p->constants)) {
xfree(p);
}
}
static size_t
autoload_i_memsize(const void *ptr)
{
return sizeof(struct autoload_data_i);
}
static const rb_data_type_t autoload_data_i_type = {
"autoload_i",
{autoload_i_mark, autoload_i_free, autoload_i_memsize, autoload_i_compact},
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
0, 0, RUBY_TYPED_FREE_IMMEDIATELY
};
static void
autoload_c_compact(void *ptr)
{
struct autoload_const *ac = ptr;
ac->mod = rb_gc_location(ac->mod);
ac->ad = rb_gc_location(ac->ad);
ac->value = rb_gc_location(ac->value);
ac->file = rb_gc_location(ac->file);
}
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
static void
autoload_c_mark(void *ptr)
{
struct autoload_const *ac = ptr;
rb_gc_mark_movable(ac->mod);
rb_gc_mark_movable(ac->ad);
rb_gc_mark_movable(ac->value);
rb_gc_mark_movable(ac->file);
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
}
static void
autoload_c_free(void *ptr)
{
struct autoload_const *ac = ptr;
list_del(&ac->cnode);
xfree(ac);
}
static size_t
autoload_c_memsize(const void *ptr)
{
return sizeof(struct autoload_const);
}
static const rb_data_type_t autoload_const_type = {
"autoload_const",
{autoload_c_mark, autoload_c_free, autoload_c_memsize, autoload_c_compact,},
0, 0, RUBY_TYPED_FREE_IMMEDIATELY
};
static struct autoload_data_i *
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
get_autoload_data(VALUE acv, struct autoload_const **acp)
{
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
struct autoload_const *ac = rb_check_typeddata(acv, &autoload_const_type);
struct autoload_data_i *ele;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
ele = rb_check_typeddata(ac->ad, &autoload_data_i_type);
/* do not reach across stack for ->state after forking: */
if (ele && ele->state && ele->fork_gen != GET_VM()->fork_gen) {
ele->state = 0;
ele->fork_gen = 0;
}
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
if (acp) *acp = ac;
return ele;
}
RUBY_FUNC_EXPORTED void
rb_autoload(VALUE mod, ID id, const char *file)
{
if (!file || !*file) {
rb_raise(rb_eArgError, "empty file name");
}
rb_autoload_str(mod, id, rb_fstring_cstr(file));
}
void
rb_autoload_str(VALUE mod, ID id, VALUE file)
{
st_data_t av;
VALUE ad;
struct st_table *tbl;
struct autoload_data_i *ele;
rb_const_entry_t *ce;
if (!rb_is_const_id(id)) {
rb_raise(rb_eNameError, "autoload must be constant name: %"PRIsVALUE"",
QUOTE_ID(id));
}
Check_Type(file, T_STRING);
if (!RSTRING_LEN(file)) {
rb_raise(rb_eArgError, "empty file name");
}
ce = rb_const_lookup(mod, id);
if (ce && ce->value != Qundef) {
return;
}
rb_const_set(mod, id, Qundef);
tbl = RCLASS_IV_TBL(mod);
if (tbl && st_lookup(tbl, (st_data_t)autoload, &av)) {
tbl = check_autoload_table((VALUE)av);
}
else {
if (!tbl) tbl = RCLASS_IV_TBL(mod) = st_init_numtable();
av = (st_data_t)TypedData_Wrap_Struct(0, &autoload_data_type, 0);
st_add_direct(tbl, (st_data_t)autoload, av);
RB_OBJ_WRITTEN(mod, Qnil, av);
DATA_PTR(av) = tbl = st_init_numtable();
}
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
file = rb_fstring(file);
if (!autoload_featuremap) {
autoload_featuremap = rb_ident_hash_new();
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
rb_obj_hide(autoload_featuremap);
rb_gc_register_mark_object(autoload_featuremap);
}
ad = rb_hash_aref(autoload_featuremap, file);
if (NIL_P(ad)) {
ad = TypedData_Make_Struct(0, struct autoload_data_i,
&autoload_data_i_type, ele);
ele->feature = file;
ele->state = 0;
list_head_init(&ele->constants);
rb_hash_aset(autoload_featuremap, file, ad);
}
else {
ele = rb_check_typeddata(ad, &autoload_data_i_type);
}
{
VALUE acv;
struct autoload_const *ac;
acv = TypedData_Make_Struct(0, struct autoload_const,
&autoload_const_type, ac);
ac->mod = mod;
ac->id = id;
ac->value = Qundef;
ac->flag = CONST_PUBLIC;
ac->ad = ad;
list_add_tail(&ele->constants, &ac->cnode);
st_insert(tbl, (st_data_t)id, (st_data_t)acv);
}
}
static void
autoload_delete(VALUE mod, ID id)
{
st_data_t val, load = 0, n = id;
if (st_lookup(RCLASS_IV_TBL(mod), (st_data_t)autoload, &val)) {
struct st_table *tbl = check_autoload_table((VALUE)val);
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
struct autoload_data_i *ele;
struct autoload_const *ac;
st_delete(tbl, &n, &load);
/* Qfalse can indicate already deleted */
if (load != Qfalse) {
ele = get_autoload_data((VALUE)load, &ac);
VM_ASSERT(ele);
if (ele) {
VM_ASSERT(!list_empty(&ele->constants));
}
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
/*
* we must delete here to avoid "already initialized" warnings
* with parallel autoload. Using list_del_init here so list_del
* works in autoload_c_free
*/
list_del_init(&ac->cnode);
variable.c: fix multiple autoload with identical file We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63387 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 04:19:20 +03:00
if (tbl->num_entries == 0) {
n = autoload;
st_delete(RCLASS_IV_TBL(mod), &n, &val);
}
}
}
}
static VALUE
check_autoload_required(VALUE mod, ID id, const char **loadingpath)
{
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
VALUE file;
VALUE load = autoload_data(mod, id);
struct autoload_data_i *ele;
const char *loading;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
if (!load || !(ele = get_autoload_data(load, 0))) {
return 0;
}
file = ele->feature;
Check_Type(file, T_STRING);
if (!RSTRING_LEN(file) || !*RSTRING_PTR(file)) {
rb_raise(rb_eArgError, "empty file name");
}
/*
* if somebody else is autoloading, we MUST wait for them, since
* rb_provide_feature can provide a feature before autoload_const_set
* completes. We must wait until autoload_const_set finishes in
* the other thread.
*/
if (ele->state && ele->state->thread != rb_thread_current()) {
return load;
}
loading = RSTRING_PTR(file);
if (!rb_feature_provided(loading, &loading)) {
return load;
}
if (loadingpath && loading) {
*loadingpath = loading;
return load;
}
return 0;
}
2019-06-22 20:49:39 +03:00
static struct autoload_const *autoloading_const_entry(VALUE mod, ID id);
2018-12-13 12:12:07 +03:00
mjit_compile.c: merge initial JIT compiler which has been developed by Takashi Kokubun <takashikkbn@gmail> as YARV-MJIT. Many of its bugs are fixed by wanabe <s.wanabe@gmail.com>. This JIT compiler is designed to be a safe migration path to introduce JIT compiler to MRI. So this commit does not include any bytecode changes or dynamic instruction modifications, which are done in original MJIT. This commit even strips off some aggressive optimizations from YARV-MJIT, and thus it's slower than YARV-MJIT too. But it's still fairly faster than Ruby 2.5 in some benchmarks (attached below). Note that this JIT compiler passes `make test`, `make test-all`, `make test-spec` without JIT, and even with JIT. Not only it's perfectly safe with JIT disabled because it does not replace VM instructions unlike MJIT, but also with JIT enabled it stably runs Ruby applications including Rails applications. I'm expecting this version as just "initial" JIT compiler. I have many optimization ideas which are skipped for initial merging, and you may easily replace this JIT compiler with a faster one by just replacing mjit_compile.c. `mjit_compile` interface is designed for the purpose. common.mk: update dependencies for mjit_compile.c. internal.h: declare `rb_vm_insn_addr2insn` for MJIT. vm.c: exclude some definitions if `-DMJIT_HEADER` is provided to compiler. This avoids to include some functions which take a long time to compile, e.g. vm_exec_core. Some of the purpose is achieved in transform_mjit_header.rb (see `IGNORED_FUNCTIONS`) but others are manually resolved for now. Load mjit_helper.h for MJIT header. mjit_helper.h: New. This is a file used only by JIT-ed code. I'll refactor `mjit_call_cfunc` later. vm_eval.c: add some #ifdef switches to skip compiling some functions like Init_vm_eval. win32/mkexports.rb: export thread/ec functions, which are used by MJIT. include/ruby/defines.h: add MJIT_FUNC_EXPORTED macro alis to clarify that a function is exported only for MJIT. array.c: export a function used by MJIT. bignum.c: ditto. class.c: ditto. compile.c: ditto. error.c: ditto. gc.c: ditto. hash.c: ditto. iseq.c: ditto. numeric.c: ditto. object.c: ditto. proc.c: ditto. re.c: ditto. st.c: ditto. string.c: ditto. thread.c: ditto. variable.c: ditto. vm_backtrace.c: ditto. vm_insnhelper.c: ditto. vm_method.c: ditto. I would like to improve maintainability of function exports, but I believe this way is acceptable as initial merging if we clarify the new exports are for MJIT (so that we can use them as TODO list to fix) and add unit tests to detect unresolved symbols. I'll add unit tests of JIT compilations in succeeding commits. Author: Takashi Kokubun <takashikkbn@gmail.com> Contributor: wanabe <s.wanabe@gmail.com> Part of [Feature #14235] --- * Known issues * Code generated by gcc is faster than clang. The benchmark may be worse in macOS. Following benchmark result is provided by gcc w/ Linux. * Performance is decreased when Google Chrome is running * JIT can work on MinGW, but it doesn't improve performance at least in short running benchmark. * Currently it doesn't perform well with Rails. We'll try to fix this before release. --- * Benchmark reslts Benchmarked with: Intel 4.0GHz i7-4790K with 16GB memory under x86-64 Ubuntu 8 Cores - 2.0.0-p0: Ruby 2.0.0-p0 - r62186: Ruby trunk (early 2.6.0), before MJIT changes - JIT off: On this commit, but without `--jit` option - JIT on: On this commit, and with `--jit` option ** Optcarrot fps Benchmark: https://github.com/mame/optcarrot | |2.0.0-p0 |r62186 |JIT off |JIT on | |:--------|:--------|:--------|:--------|:--------| |fps |37.32 |51.46 |51.31 |58.88 | |vs 2.0.0 |1.00x |1.38x |1.37x |1.58x | ** MJIT benchmarks Benchmark: https://github.com/benchmark-driver/mjit-benchmarks (Original: https://github.com/vnmakarov/ruby/tree/rtl_mjit_branch/MJIT-benchmarks) | |2.0.0-p0 |r62186 |JIT off |JIT on | |:----------|:--------|:--------|:--------|:--------| |aread |1.00 |1.09 |1.07 |2.19 | |aref |1.00 |1.13 |1.11 |2.22 | |aset |1.00 |1.50 |1.45 |2.64 | |awrite |1.00 |1.17 |1.13 |2.20 | |call |1.00 |1.29 |1.26 |2.02 | |const2 |1.00 |1.10 |1.10 |2.19 | |const |1.00 |1.11 |1.10 |2.19 | |fannk |1.00 |1.04 |1.02 |1.00 | |fib |1.00 |1.32 |1.31 |1.84 | |ivread |1.00 |1.13 |1.12 |2.43 | |ivwrite |1.00 |1.23 |1.21 |2.40 | |mandelbrot |1.00 |1.13 |1.16 |1.28 | |meteor |1.00 |2.97 |2.92 |3.17 | |nbody |1.00 |1.17 |1.15 |1.49 | |nest-ntimes|1.00 |1.22 |1.20 |1.39 | |nest-while |1.00 |1.10 |1.10 |1.37 | |norm |1.00 |1.18 |1.16 |1.24 | |nsvb |1.00 |1.16 |1.16 |1.17 | |red-black |1.00 |1.02 |0.99 |1.12 | |sieve |1.00 |1.30 |1.28 |1.62 | |trees |1.00 |1.14 |1.13 |1.19 | |while |1.00 |1.12 |1.11 |2.41 | ** Discourse's script/bench.rb Benchmark: https://github.com/discourse/discourse/blob/v1.8.7/script/bench.rb NOTE: Rails performance was somehow a little degraded with JIT for now. We should fix this. (At least I know opt_aref is performing badly in JIT and I have an idea to fix it. Please wait for the fix.) *** JIT off Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 17 75: 18 90: 22 99: 29 home_admin: 50: 21 75: 21 90: 27 99: 40 topic_admin: 50: 17 75: 18 90: 22 99: 32 categories: 50: 35 75: 41 90: 43 99: 77 home: 50: 39 75: 46 90: 49 99: 95 topic: 50: 46 75: 52 90: 56 99: 101 *** JIT on Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 19 75: 21 90: 25 99: 33 home_admin: 50: 24 75: 26 90: 30 99: 35 topic_admin: 50: 19 75: 20 90: 25 99: 30 categories: 50: 40 75: 44 90: 48 99: 76 home: 50: 42 75: 48 90: 51 99: 89 topic: 50: 49 75: 55 90: 58 99: 99 git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62197 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-02-04 14:22:28 +03:00
MJIT_FUNC_EXPORTED int
rb_autoloading_value(VALUE mod, ID id, VALUE* value, rb_const_flag_t *flag)
2018-12-13 12:12:07 +03:00
{
struct autoload_const *ac = autoloading_const_entry(mod, id);
if (!ac) return FALSE;
if (value) {
*value = ac->value;
}
if (flag) {
*flag = ac->flag;
}
return TRUE;
}
struct autoload_const *
autoloading_const_entry(VALUE mod, ID id)
{
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
VALUE load = autoload_data(mod, id);
struct autoload_data_i *ele;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
struct autoload_const *ac;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
if (!load || !(ele = get_autoload_data(load, &ac))) {
return 0;
}
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
if (ele->state && ele->state->thread == rb_thread_current()) {
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
if (ac->value != Qundef) {
2018-12-13 12:12:07 +03:00
return ac;
}
}
return 0;
}
static int
autoload_defined_p(VALUE mod, ID id)
{
rb_const_entry_t *ce = rb_const_lookup(mod, id);
if (!ce || ce->value != Qundef) {
return 0;
}
return !rb_autoloading_value(mod, id, NULL, NULL);
}
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
static void const_tbl_update(struct autoload_const *);
static VALUE
2020-04-07 14:20:34 +03:00
autoload_const_set(struct autoload_const *ac)
{
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
VALUE klass = ac->mod;
ID id = ac->id;
check_before_mod_set(klass, id, ac->value, "constant");
RB_VM_LOCK_ENTER();
{
const_tbl_update(ac);
}
RB_VM_LOCK_LEAVE();
return 0; /* ignored */
}
static VALUE
autoload_require(VALUE arg)
{
struct autoload_state *state = (struct autoload_state *)arg;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
struct autoload_const *ac = state->ac;
struct autoload_data_i *ele;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
ele = rb_check_typeddata(ac->ad, &autoload_data_i_type);
/* this may release GVL and switch threads: */
state->result = rb_funcall(rb_vm_top_self(), rb_intern("require"), 1,
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
ele->feature);
return state->result;
}
static VALUE
autoload_reset(VALUE arg)
{
struct autoload_state *state = (struct autoload_state *)arg;
int need_wakeups = 0;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
struct autoload_const *ac = state->ac;
struct autoload_data_i *ele;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
ele = rb_check_typeddata(ac->ad, &autoload_data_i_type);
if (ele->state == state) {
need_wakeups = 1;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
ele->state = 0;
ele->fork_gen = 0;
}
/* At the last, move a value defined in autoload to constant table */
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
if (RTEST(state->result)) {
struct autoload_const *next;
list_for_each_safe(&ele->constants, ac, next, cnode) {
if (ac->value != Qundef) {
2020-04-07 14:20:34 +03:00
autoload_const_set(ac);
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
}
}
}
/* wakeup any waiters we had */
if (need_wakeups) {
struct autoload_state *cur = 0, *nxt;
2021-06-23 04:50:09 +03:00
list_for_each_safe(&state->waitq, cur, nxt, waitq.n) {
VALUE th = cur->thread;
cur->thread = Qfalse;
2021-06-23 04:50:09 +03:00
list_del_init(&cur->waitq.n); /* idempotent */
/*
* cur is stored on the stack of cur->waiting_th,
* do not touch cur after waking up waiting_th
*/
rb_thread_wakeup_alive(th);
}
}
return 0; /* ignored */
}
static VALUE
autoload_sleep(VALUE arg)
{
struct autoload_state *state = (struct autoload_state *)arg;
/*
* autoload_reset in other thread will resume us and remove us
* from the waitq list
*/
do {
rb_thread_sleep_deadly();
} while (state->thread != Qfalse);
return Qfalse;
}
static VALUE
autoload_sleep_done(VALUE arg)
{
struct autoload_state *state = (struct autoload_state *)arg;
if (state->thread != Qfalse && rb_thread_to_be_killed(state->thread)) {
2021-06-23 04:50:09 +03:00
list_del(&state->waitq.n); /* idempotent after list_del_init */
}
return Qfalse;
}
VALUE
rb_autoload_load(VALUE mod, ID id)
{
VALUE load, result;
const char *loading = 0, *src;
struct autoload_data_i *ele;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
struct autoload_const *ac;
struct autoload_state state;
int flag = -1;
rb_const_entry_t *ce;
if (!autoload_defined_p(mod, id)) return Qfalse;
load = check_autoload_required(mod, id, &loading);
if (!load) return Qfalse;
src = rb_sourcefile();
if (src && loading && strcmp(src, loading) == 0) return Qfalse;
if ((ce = rb_const_lookup(mod, id))) {
flag = ce->flag & (CONST_DEPRECATED | CONST_VISIBILITY_MASK);
}
/* set ele->state for a marker of autoloading thread */
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
if (!(ele = get_autoload_data(load, &ac))) {
return Qfalse;
}
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
state.ac = ac;
state.thread = rb_thread_current();
if (!ele->state) {
ele->state = &state;
ele->fork_gen = GET_VM()->fork_gen;
/*
* autoload_reset will wake up any threads added to this
* if and only if the GVL is released during autoload_require
*/
2021-06-23 04:50:09 +03:00
list_head_init(&state.waitq);
}
else if (state.thread == ele->state->thread) {
return Qfalse;
}
else {
2021-06-23 04:50:09 +03:00
list_add_tail(&ele->state->waitq, &state.waitq.n);
rb_ensure(autoload_sleep, (VALUE)&state,
autoload_sleep_done, (VALUE)&state);
}
/* autoload_data_i can be deleted by another thread while require */
state.result = Qfalse;
result = rb_ensure(autoload_require, (VALUE)&state,
autoload_reset, (VALUE)&state);
if (!(ce = rb_const_lookup(mod, id)) || ce->value == Qundef) {
rb_const_remove(mod, id);
}
else if (flag > 0) {
ce->flag |= flag;
}
RB_GC_GUARD(load);
return result;
}
VALUE
rb_autoload_p(VALUE mod, ID id)
{
return rb_autoload_at_p(mod, id, TRUE);
}
VALUE
rb_autoload_at_p(VALUE mod, ID id, int recur)
{
VALUE load;
struct autoload_data_i *ele;
while (!autoload_defined_p(mod, id)) {
2019-06-21 11:29:08 +03:00
if (!recur) return Qnil;
mod = RCLASS_SUPER(mod);
if (!mod) return Qnil;
}
load = check_autoload_required(mod, id, 0);
if (!load) return Qnil;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
return (ele = get_autoload_data(load, 0)) ? ele->feature : Qnil;
}
mjit_compile.c: merge initial JIT compiler which has been developed by Takashi Kokubun <takashikkbn@gmail> as YARV-MJIT. Many of its bugs are fixed by wanabe <s.wanabe@gmail.com>. This JIT compiler is designed to be a safe migration path to introduce JIT compiler to MRI. So this commit does not include any bytecode changes or dynamic instruction modifications, which are done in original MJIT. This commit even strips off some aggressive optimizations from YARV-MJIT, and thus it's slower than YARV-MJIT too. But it's still fairly faster than Ruby 2.5 in some benchmarks (attached below). Note that this JIT compiler passes `make test`, `make test-all`, `make test-spec` without JIT, and even with JIT. Not only it's perfectly safe with JIT disabled because it does not replace VM instructions unlike MJIT, but also with JIT enabled it stably runs Ruby applications including Rails applications. I'm expecting this version as just "initial" JIT compiler. I have many optimization ideas which are skipped for initial merging, and you may easily replace this JIT compiler with a faster one by just replacing mjit_compile.c. `mjit_compile` interface is designed for the purpose. common.mk: update dependencies for mjit_compile.c. internal.h: declare `rb_vm_insn_addr2insn` for MJIT. vm.c: exclude some definitions if `-DMJIT_HEADER` is provided to compiler. This avoids to include some functions which take a long time to compile, e.g. vm_exec_core. Some of the purpose is achieved in transform_mjit_header.rb (see `IGNORED_FUNCTIONS`) but others are manually resolved for now. Load mjit_helper.h for MJIT header. mjit_helper.h: New. This is a file used only by JIT-ed code. I'll refactor `mjit_call_cfunc` later. vm_eval.c: add some #ifdef switches to skip compiling some functions like Init_vm_eval. win32/mkexports.rb: export thread/ec functions, which are used by MJIT. include/ruby/defines.h: add MJIT_FUNC_EXPORTED macro alis to clarify that a function is exported only for MJIT. array.c: export a function used by MJIT. bignum.c: ditto. class.c: ditto. compile.c: ditto. error.c: ditto. gc.c: ditto. hash.c: ditto. iseq.c: ditto. numeric.c: ditto. object.c: ditto. proc.c: ditto. re.c: ditto. st.c: ditto. string.c: ditto. thread.c: ditto. variable.c: ditto. vm_backtrace.c: ditto. vm_insnhelper.c: ditto. vm_method.c: ditto. I would like to improve maintainability of function exports, but I believe this way is acceptable as initial merging if we clarify the new exports are for MJIT (so that we can use them as TODO list to fix) and add unit tests to detect unresolved symbols. I'll add unit tests of JIT compilations in succeeding commits. Author: Takashi Kokubun <takashikkbn@gmail.com> Contributor: wanabe <s.wanabe@gmail.com> Part of [Feature #14235] --- * Known issues * Code generated by gcc is faster than clang. The benchmark may be worse in macOS. Following benchmark result is provided by gcc w/ Linux. * Performance is decreased when Google Chrome is running * JIT can work on MinGW, but it doesn't improve performance at least in short running benchmark. * Currently it doesn't perform well with Rails. We'll try to fix this before release. --- * Benchmark reslts Benchmarked with: Intel 4.0GHz i7-4790K with 16GB memory under x86-64 Ubuntu 8 Cores - 2.0.0-p0: Ruby 2.0.0-p0 - r62186: Ruby trunk (early 2.6.0), before MJIT changes - JIT off: On this commit, but without `--jit` option - JIT on: On this commit, and with `--jit` option ** Optcarrot fps Benchmark: https://github.com/mame/optcarrot | |2.0.0-p0 |r62186 |JIT off |JIT on | |:--------|:--------|:--------|:--------|:--------| |fps |37.32 |51.46 |51.31 |58.88 | |vs 2.0.0 |1.00x |1.38x |1.37x |1.58x | ** MJIT benchmarks Benchmark: https://github.com/benchmark-driver/mjit-benchmarks (Original: https://github.com/vnmakarov/ruby/tree/rtl_mjit_branch/MJIT-benchmarks) | |2.0.0-p0 |r62186 |JIT off |JIT on | |:----------|:--------|:--------|:--------|:--------| |aread |1.00 |1.09 |1.07 |2.19 | |aref |1.00 |1.13 |1.11 |2.22 | |aset |1.00 |1.50 |1.45 |2.64 | |awrite |1.00 |1.17 |1.13 |2.20 | |call |1.00 |1.29 |1.26 |2.02 | |const2 |1.00 |1.10 |1.10 |2.19 | |const |1.00 |1.11 |1.10 |2.19 | |fannk |1.00 |1.04 |1.02 |1.00 | |fib |1.00 |1.32 |1.31 |1.84 | |ivread |1.00 |1.13 |1.12 |2.43 | |ivwrite |1.00 |1.23 |1.21 |2.40 | |mandelbrot |1.00 |1.13 |1.16 |1.28 | |meteor |1.00 |2.97 |2.92 |3.17 | |nbody |1.00 |1.17 |1.15 |1.49 | |nest-ntimes|1.00 |1.22 |1.20 |1.39 | |nest-while |1.00 |1.10 |1.10 |1.37 | |norm |1.00 |1.18 |1.16 |1.24 | |nsvb |1.00 |1.16 |1.16 |1.17 | |red-black |1.00 |1.02 |0.99 |1.12 | |sieve |1.00 |1.30 |1.28 |1.62 | |trees |1.00 |1.14 |1.13 |1.19 | |while |1.00 |1.12 |1.11 |2.41 | ** Discourse's script/bench.rb Benchmark: https://github.com/discourse/discourse/blob/v1.8.7/script/bench.rb NOTE: Rails performance was somehow a little degraded with JIT for now. We should fix this. (At least I know opt_aref is performing badly in JIT and I have an idea to fix it. Please wait for the fix.) *** JIT off Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 17 75: 18 90: 22 99: 29 home_admin: 50: 21 75: 21 90: 27 99: 40 topic_admin: 50: 17 75: 18 90: 22 99: 32 categories: 50: 35 75: 41 90: 43 99: 77 home: 50: 39 75: 46 90: 49 99: 95 topic: 50: 46 75: 52 90: 56 99: 101 *** JIT on Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 19 75: 21 90: 25 99: 33 home_admin: 50: 24 75: 26 90: 30 99: 35 topic_admin: 50: 19 75: 20 90: 25 99: 30 categories: 50: 40 75: 44 90: 48 99: 76 home: 50: 42 75: 48 90: 51 99: 89 topic: 50: 49 75: 55 90: 58 99: 99 git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62197 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-02-04 14:22:28 +03:00
MJIT_FUNC_EXPORTED void
rb_const_warn_if_deprecated(const rb_const_entry_t *ce, VALUE klass, ID id)
{
if (RB_CONST_DEPRECATED_P(ce) &&
rb_warning_category_enabled_p(RB_WARN_CATEGORY_DEPRECATED)) {
if (klass == rb_cObject) {
rb_category_warn(RB_WARN_CATEGORY_DEPRECATED, "constant ::%"PRIsVALUE" is deprecated", QUOTE_ID(id));
}
else {
rb_category_warn(RB_WARN_CATEGORY_DEPRECATED, "constant %"PRIsVALUE"::%"PRIsVALUE" is deprecated",
rb_class_name(klass), QUOTE_ID(id));
}
}
}
static VALUE
rb_const_get_0(VALUE klass, ID id, int exclude, int recurse, int visibility)
{
VALUE c = rb_const_search(klass, id, exclude, recurse, visibility);
if (c != Qundef) {
if (UNLIKELY(!rb_ractor_main_p())) {
if (!rb_ractor_shareable_p(c)) {
rb_raise(rb_eRactorIsolationError, "can not access non-shareable objects in constant %"PRIsVALUE"::%s by non-main Ractor.", rb_class_path(klass), rb_id2name(id));
}
}
return c;
}
return rb_const_missing(klass, ID2SYM(id));
}
static VALUE
2018-12-13 12:27:32 +03:00
rb_const_search_from(VALUE klass, ID id, int exclude, int recurse, int visibility)
{
VALUE value, current;
bool first_iteration = true;
for (current = klass;
RTEST(current);
current = RCLASS_SUPER(current), first_iteration = false) {
VALUE tmp;
VALUE am = 0;
rb_const_entry_t *ce;
if (!first_iteration && RCLASS_ORIGIN(current) != current) {
// This item in the super chain has an origin iclass
// that comes later in the chain. Skip this item so
// prepended modules take precedence.
continue;
}
// Do lookup in original class or module in case we are at an origin
// iclass in the chain.
tmp = current;
if (BUILTIN_TYPE(tmp) == T_ICLASS) tmp = RBASIC(tmp)->klass;
// Do the lookup. Loop in case of autoload.
while ((ce = rb_const_lookup(tmp, id))) {
if (visibility && RB_CONST_PRIVATE_P(ce)) {
GET_EC()->private_const_reference = tmp;
return Qundef;
}
rb_const_warn_if_deprecated(ce, tmp, id);
value = ce->value;
if (value == Qundef) {
2018-12-13 12:12:07 +03:00
struct autoload_const *ac;
if (am == tmp) break;
am = tmp;
2018-12-13 12:12:07 +03:00
ac = autoloading_const_entry(tmp, id);
if (ac) return ac->value;
rb_autoload_load(tmp, id);
continue;
}
2019-06-22 19:47:40 +03:00
if (exclude && tmp == rb_cObject) {
goto not_found;
}
return value;
}
if (!recurse) break;
}
not_found:
GET_EC()->private_const_reference = 0;
return Qundef;
}
2018-12-13 12:27:32 +03:00
static VALUE
rb_const_search(VALUE klass, ID id, int exclude, int recurse, int visibility)
{
VALUE value;
if (klass == rb_cObject) exclude = FALSE;
value = rb_const_search_from(klass, id, exclude, recurse, visibility);
if (value != Qundef) return value;
if (exclude) return value;
if (BUILTIN_TYPE(klass) != T_MODULE) return value;
/* search global const too, if klass is a module */
return rb_const_search_from(rb_cObject, id, FALSE, recurse, visibility);
}
VALUE
rb_const_get_from(VALUE klass, ID id)
{
return rb_const_get_0(klass, id, TRUE, TRUE, FALSE);
}
VALUE
rb_const_get(VALUE klass, ID id)
{
return rb_const_get_0(klass, id, FALSE, TRUE, FALSE);
}
VALUE
rb_const_get_at(VALUE klass, ID id)
{
return rb_const_get_0(klass, id, TRUE, FALSE, FALSE);
}
mjit_compile.c: merge initial JIT compiler which has been developed by Takashi Kokubun <takashikkbn@gmail> as YARV-MJIT. Many of its bugs are fixed by wanabe <s.wanabe@gmail.com>. This JIT compiler is designed to be a safe migration path to introduce JIT compiler to MRI. So this commit does not include any bytecode changes or dynamic instruction modifications, which are done in original MJIT. This commit even strips off some aggressive optimizations from YARV-MJIT, and thus it's slower than YARV-MJIT too. But it's still fairly faster than Ruby 2.5 in some benchmarks (attached below). Note that this JIT compiler passes `make test`, `make test-all`, `make test-spec` without JIT, and even with JIT. Not only it's perfectly safe with JIT disabled because it does not replace VM instructions unlike MJIT, but also with JIT enabled it stably runs Ruby applications including Rails applications. I'm expecting this version as just "initial" JIT compiler. I have many optimization ideas which are skipped for initial merging, and you may easily replace this JIT compiler with a faster one by just replacing mjit_compile.c. `mjit_compile` interface is designed for the purpose. common.mk: update dependencies for mjit_compile.c. internal.h: declare `rb_vm_insn_addr2insn` for MJIT. vm.c: exclude some definitions if `-DMJIT_HEADER` is provided to compiler. This avoids to include some functions which take a long time to compile, e.g. vm_exec_core. Some of the purpose is achieved in transform_mjit_header.rb (see `IGNORED_FUNCTIONS`) but others are manually resolved for now. Load mjit_helper.h for MJIT header. mjit_helper.h: New. This is a file used only by JIT-ed code. I'll refactor `mjit_call_cfunc` later. vm_eval.c: add some #ifdef switches to skip compiling some functions like Init_vm_eval. win32/mkexports.rb: export thread/ec functions, which are used by MJIT. include/ruby/defines.h: add MJIT_FUNC_EXPORTED macro alis to clarify that a function is exported only for MJIT. array.c: export a function used by MJIT. bignum.c: ditto. class.c: ditto. compile.c: ditto. error.c: ditto. gc.c: ditto. hash.c: ditto. iseq.c: ditto. numeric.c: ditto. object.c: ditto. proc.c: ditto. re.c: ditto. st.c: ditto. string.c: ditto. thread.c: ditto. variable.c: ditto. vm_backtrace.c: ditto. vm_insnhelper.c: ditto. vm_method.c: ditto. I would like to improve maintainability of function exports, but I believe this way is acceptable as initial merging if we clarify the new exports are for MJIT (so that we can use them as TODO list to fix) and add unit tests to detect unresolved symbols. I'll add unit tests of JIT compilations in succeeding commits. Author: Takashi Kokubun <takashikkbn@gmail.com> Contributor: wanabe <s.wanabe@gmail.com> Part of [Feature #14235] --- * Known issues * Code generated by gcc is faster than clang. The benchmark may be worse in macOS. Following benchmark result is provided by gcc w/ Linux. * Performance is decreased when Google Chrome is running * JIT can work on MinGW, but it doesn't improve performance at least in short running benchmark. * Currently it doesn't perform well with Rails. We'll try to fix this before release. --- * Benchmark reslts Benchmarked with: Intel 4.0GHz i7-4790K with 16GB memory under x86-64 Ubuntu 8 Cores - 2.0.0-p0: Ruby 2.0.0-p0 - r62186: Ruby trunk (early 2.6.0), before MJIT changes - JIT off: On this commit, but without `--jit` option - JIT on: On this commit, and with `--jit` option ** Optcarrot fps Benchmark: https://github.com/mame/optcarrot | |2.0.0-p0 |r62186 |JIT off |JIT on | |:--------|:--------|:--------|:--------|:--------| |fps |37.32 |51.46 |51.31 |58.88 | |vs 2.0.0 |1.00x |1.38x |1.37x |1.58x | ** MJIT benchmarks Benchmark: https://github.com/benchmark-driver/mjit-benchmarks (Original: https://github.com/vnmakarov/ruby/tree/rtl_mjit_branch/MJIT-benchmarks) | |2.0.0-p0 |r62186 |JIT off |JIT on | |:----------|:--------|:--------|:--------|:--------| |aread |1.00 |1.09 |1.07 |2.19 | |aref |1.00 |1.13 |1.11 |2.22 | |aset |1.00 |1.50 |1.45 |2.64 | |awrite |1.00 |1.17 |1.13 |2.20 | |call |1.00 |1.29 |1.26 |2.02 | |const2 |1.00 |1.10 |1.10 |2.19 | |const |1.00 |1.11 |1.10 |2.19 | |fannk |1.00 |1.04 |1.02 |1.00 | |fib |1.00 |1.32 |1.31 |1.84 | |ivread |1.00 |1.13 |1.12 |2.43 | |ivwrite |1.00 |1.23 |1.21 |2.40 | |mandelbrot |1.00 |1.13 |1.16 |1.28 | |meteor |1.00 |2.97 |2.92 |3.17 | |nbody |1.00 |1.17 |1.15 |1.49 | |nest-ntimes|1.00 |1.22 |1.20 |1.39 | |nest-while |1.00 |1.10 |1.10 |1.37 | |norm |1.00 |1.18 |1.16 |1.24 | |nsvb |1.00 |1.16 |1.16 |1.17 | |red-black |1.00 |1.02 |0.99 |1.12 | |sieve |1.00 |1.30 |1.28 |1.62 | |trees |1.00 |1.14 |1.13 |1.19 | |while |1.00 |1.12 |1.11 |2.41 | ** Discourse's script/bench.rb Benchmark: https://github.com/discourse/discourse/blob/v1.8.7/script/bench.rb NOTE: Rails performance was somehow a little degraded with JIT for now. We should fix this. (At least I know opt_aref is performing badly in JIT and I have an idea to fix it. Please wait for the fix.) *** JIT off Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 17 75: 18 90: 22 99: 29 home_admin: 50: 21 75: 21 90: 27 99: 40 topic_admin: 50: 17 75: 18 90: 22 99: 32 categories: 50: 35 75: 41 90: 43 99: 77 home: 50: 39 75: 46 90: 49 99: 95 topic: 50: 46 75: 52 90: 56 99: 101 *** JIT on Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 19 75: 21 90: 25 99: 33 home_admin: 50: 24 75: 26 90: 30 99: 35 topic_admin: 50: 19 75: 20 90: 25 99: 30 categories: 50: 40 75: 44 90: 48 99: 76 home: 50: 42 75: 48 90: 51 99: 89 topic: 50: 49 75: 55 90: 58 99: 99 git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62197 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-02-04 14:22:28 +03:00
MJIT_FUNC_EXPORTED VALUE
rb_public_const_get_from(VALUE klass, ID id)
{
return rb_const_get_0(klass, id, TRUE, TRUE, TRUE);
}
mjit_compile.c: merge initial JIT compiler which has been developed by Takashi Kokubun <takashikkbn@gmail> as YARV-MJIT. Many of its bugs are fixed by wanabe <s.wanabe@gmail.com>. This JIT compiler is designed to be a safe migration path to introduce JIT compiler to MRI. So this commit does not include any bytecode changes or dynamic instruction modifications, which are done in original MJIT. This commit even strips off some aggressive optimizations from YARV-MJIT, and thus it's slower than YARV-MJIT too. But it's still fairly faster than Ruby 2.5 in some benchmarks (attached below). Note that this JIT compiler passes `make test`, `make test-all`, `make test-spec` without JIT, and even with JIT. Not only it's perfectly safe with JIT disabled because it does not replace VM instructions unlike MJIT, but also with JIT enabled it stably runs Ruby applications including Rails applications. I'm expecting this version as just "initial" JIT compiler. I have many optimization ideas which are skipped for initial merging, and you may easily replace this JIT compiler with a faster one by just replacing mjit_compile.c. `mjit_compile` interface is designed for the purpose. common.mk: update dependencies for mjit_compile.c. internal.h: declare `rb_vm_insn_addr2insn` for MJIT. vm.c: exclude some definitions if `-DMJIT_HEADER` is provided to compiler. This avoids to include some functions which take a long time to compile, e.g. vm_exec_core. Some of the purpose is achieved in transform_mjit_header.rb (see `IGNORED_FUNCTIONS`) but others are manually resolved for now. Load mjit_helper.h for MJIT header. mjit_helper.h: New. This is a file used only by JIT-ed code. I'll refactor `mjit_call_cfunc` later. vm_eval.c: add some #ifdef switches to skip compiling some functions like Init_vm_eval. win32/mkexports.rb: export thread/ec functions, which are used by MJIT. include/ruby/defines.h: add MJIT_FUNC_EXPORTED macro alis to clarify that a function is exported only for MJIT. array.c: export a function used by MJIT. bignum.c: ditto. class.c: ditto. compile.c: ditto. error.c: ditto. gc.c: ditto. hash.c: ditto. iseq.c: ditto. numeric.c: ditto. object.c: ditto. proc.c: ditto. re.c: ditto. st.c: ditto. string.c: ditto. thread.c: ditto. variable.c: ditto. vm_backtrace.c: ditto. vm_insnhelper.c: ditto. vm_method.c: ditto. I would like to improve maintainability of function exports, but I believe this way is acceptable as initial merging if we clarify the new exports are for MJIT (so that we can use them as TODO list to fix) and add unit tests to detect unresolved symbols. I'll add unit tests of JIT compilations in succeeding commits. Author: Takashi Kokubun <takashikkbn@gmail.com> Contributor: wanabe <s.wanabe@gmail.com> Part of [Feature #14235] --- * Known issues * Code generated by gcc is faster than clang. The benchmark may be worse in macOS. Following benchmark result is provided by gcc w/ Linux. * Performance is decreased when Google Chrome is running * JIT can work on MinGW, but it doesn't improve performance at least in short running benchmark. * Currently it doesn't perform well with Rails. We'll try to fix this before release. --- * Benchmark reslts Benchmarked with: Intel 4.0GHz i7-4790K with 16GB memory under x86-64 Ubuntu 8 Cores - 2.0.0-p0: Ruby 2.0.0-p0 - r62186: Ruby trunk (early 2.6.0), before MJIT changes - JIT off: On this commit, but without `--jit` option - JIT on: On this commit, and with `--jit` option ** Optcarrot fps Benchmark: https://github.com/mame/optcarrot | |2.0.0-p0 |r62186 |JIT off |JIT on | |:--------|:--------|:--------|:--------|:--------| |fps |37.32 |51.46 |51.31 |58.88 | |vs 2.0.0 |1.00x |1.38x |1.37x |1.58x | ** MJIT benchmarks Benchmark: https://github.com/benchmark-driver/mjit-benchmarks (Original: https://github.com/vnmakarov/ruby/tree/rtl_mjit_branch/MJIT-benchmarks) | |2.0.0-p0 |r62186 |JIT off |JIT on | |:----------|:--------|:--------|:--------|:--------| |aread |1.00 |1.09 |1.07 |2.19 | |aref |1.00 |1.13 |1.11 |2.22 | |aset |1.00 |1.50 |1.45 |2.64 | |awrite |1.00 |1.17 |1.13 |2.20 | |call |1.00 |1.29 |1.26 |2.02 | |const2 |1.00 |1.10 |1.10 |2.19 | |const |1.00 |1.11 |1.10 |2.19 | |fannk |1.00 |1.04 |1.02 |1.00 | |fib |1.00 |1.32 |1.31 |1.84 | |ivread |1.00 |1.13 |1.12 |2.43 | |ivwrite |1.00 |1.23 |1.21 |2.40 | |mandelbrot |1.00 |1.13 |1.16 |1.28 | |meteor |1.00 |2.97 |2.92 |3.17 | |nbody |1.00 |1.17 |1.15 |1.49 | |nest-ntimes|1.00 |1.22 |1.20 |1.39 | |nest-while |1.00 |1.10 |1.10 |1.37 | |norm |1.00 |1.18 |1.16 |1.24 | |nsvb |1.00 |1.16 |1.16 |1.17 | |red-black |1.00 |1.02 |0.99 |1.12 | |sieve |1.00 |1.30 |1.28 |1.62 | |trees |1.00 |1.14 |1.13 |1.19 | |while |1.00 |1.12 |1.11 |2.41 | ** Discourse's script/bench.rb Benchmark: https://github.com/discourse/discourse/blob/v1.8.7/script/bench.rb NOTE: Rails performance was somehow a little degraded with JIT for now. We should fix this. (At least I know opt_aref is performing badly in JIT and I have an idea to fix it. Please wait for the fix.) *** JIT off Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 17 75: 18 90: 22 99: 29 home_admin: 50: 21 75: 21 90: 27 99: 40 topic_admin: 50: 17 75: 18 90: 22 99: 32 categories: 50: 35 75: 41 90: 43 99: 77 home: 50: 39 75: 46 90: 49 99: 95 topic: 50: 46 75: 52 90: 56 99: 101 *** JIT on Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 19 75: 21 90: 25 99: 33 home_admin: 50: 24 75: 26 90: 30 99: 35 topic_admin: 50: 19 75: 20 90: 25 99: 30 categories: 50: 40 75: 44 90: 48 99: 76 home: 50: 42 75: 48 90: 51 99: 89 topic: 50: 49 75: 55 90: 58 99: 99 git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62197 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-02-04 14:22:28 +03:00
MJIT_FUNC_EXPORTED VALUE
rb_public_const_get_at(VALUE klass, ID id)
{
return rb_const_get_0(klass, id, TRUE, FALSE, TRUE);
}
NORETURN(static void undefined_constant(VALUE mod, VALUE name));
static void
undefined_constant(VALUE mod, VALUE name)
{
rb_name_err_raise("constant %2$s::%1$s not defined",
mod, name);
}
static VALUE
rb_const_location_from(VALUE klass, ID id, int exclude, int recurse, int visibility)
{
while (RTEST(klass)) {
2019-06-22 19:47:40 +03:00
rb_const_entry_t *ce;
2019-06-22 19:47:40 +03:00
while ((ce = rb_const_lookup(klass, id))) {
if (visibility && RB_CONST_PRIVATE_P(ce)) {
return Qnil;
}
if (exclude && klass == rb_cObject) {
goto not_found;
}
if (NIL_P(ce->file)) return rb_ary_new();
return rb_assoc_new(ce->file, INT2NUM(ce->line));
}
if (!recurse) break;
klass = RCLASS_SUPER(klass);
}
not_found:
return Qnil;
}
static VALUE
rb_const_location(VALUE klass, ID id, int exclude, int recurse, int visibility)
{
VALUE loc;
if (klass == rb_cObject) exclude = FALSE;
loc = rb_const_location_from(klass, id, exclude, recurse, visibility);
if (!NIL_P(loc)) return loc;
if (exclude) return loc;
if (BUILTIN_TYPE(klass) != T_MODULE) return loc;
/* search global const too, if klass is a module */
return rb_const_location_from(rb_cObject, id, FALSE, recurse, visibility);
}
VALUE
rb_const_source_location(VALUE klass, ID id)
{
return rb_const_location(klass, id, FALSE, TRUE, FALSE);
}
MJIT_FUNC_EXPORTED VALUE
rb_const_source_location_at(VALUE klass, ID id)
{
return rb_const_location(klass, id, TRUE, FALSE, FALSE);
}
/*
* call-seq:
* remove_const(sym) -> obj
*
* Removes the definition of the given constant, returning that
* constant's previous value. If that constant referred to
* a module, this will not change that module's name and can lead
* to confusion.
*/
VALUE
rb_mod_remove_const(VALUE mod, VALUE name)
{
const ID id = id_for_var(mod, name, a, constant);
if (!id) {
undefined_constant(mod, name);
}
return rb_const_remove(mod, id);
}
VALUE
rb_const_remove(VALUE mod, ID id)
{
VALUE val;
rb_const_entry_t *ce;
rb_check_frozen(mod);
ce = rb_const_lookup(mod, id);
if (!ce || !rb_id_table_delete(RCLASS_CONST_TBL(mod), id)) {
if (rb_const_defined_at(mod, id)) {
rb_name_err_raise("cannot remove %2$s::%1$s",
mod, ID2SYM(id));
}
undefined_constant(mod, ID2SYM(id));
}
rb_clear_constant_cache();
val = ce->value;
if (val == Qundef) {
autoload_delete(mod, id);
val = Qnil;
}
xfree(ce);
return val;
}
static int
cv_i_update(st_data_t *k, st_data_t *v, st_data_t a, int existing)
{
if (existing) return ST_STOP;
*v = a;
return ST_CONTINUE;
}
static enum rb_id_table_iterator_result
sv_i(ID key, VALUE v, void *a)
{
rb_const_entry_t *ce = (rb_const_entry_t *)v;
st_table *tbl = a;
if (rb_is_const_id(key)) {
st_update(tbl, (st_data_t)key, cv_i_update, (st_data_t)ce);
}
return ID_TABLE_CONTINUE;
}
static enum rb_id_table_iterator_result
rb_local_constants_i(ID const_name, VALUE const_value, void *ary)
{
if (rb_is_const_id(const_name) && !RB_CONST_PRIVATE_P((rb_const_entry_t *)const_value)) {
rb_ary_push((VALUE)ary, ID2SYM(const_name));
}
return ID_TABLE_CONTINUE;
}
static VALUE
rb_local_constants(VALUE mod)
{
struct rb_id_table *tbl = RCLASS_CONST_TBL(mod);
VALUE ary;
if (!tbl) return rb_ary_new2(0);
RB_VM_LOCK_ENTER();
{
ary = rb_ary_new2(rb_id_table_size(tbl));
rb_id_table_foreach(tbl, rb_local_constants_i, (void *)ary);
}
RB_VM_LOCK_LEAVE();
return ary;
}
void*
rb_mod_const_at(VALUE mod, void *data)
{
st_table *tbl = data;
if (!tbl) {
tbl = st_init_numtable();
}
if (RCLASS_CONST_TBL(mod)) {
RB_VM_LOCK_ENTER();
{
rb_id_table_foreach(RCLASS_CONST_TBL(mod), sv_i, tbl);
}
RB_VM_LOCK_LEAVE();
}
return tbl;
}
void*
rb_mod_const_of(VALUE mod, void *data)
{
VALUE tmp = mod;
for (;;) {
data = rb_mod_const_at(tmp, data);
tmp = RCLASS_SUPER(tmp);
if (!tmp) break;
if (tmp == rb_cObject && mod != rb_cObject) break;
}
return data;
}
static int
list_i(st_data_t key, st_data_t value, VALUE ary)
{
ID sym = (ID)key;
rb_const_entry_t *ce = (rb_const_entry_t *)value;
if (RB_CONST_PUBLIC_P(ce)) rb_ary_push(ary, ID2SYM(sym));
return ST_CONTINUE;
}
VALUE
rb_const_list(void *data)
{
st_table *tbl = data;
VALUE ary;
if (!tbl) return rb_ary_new2(0);
ary = rb_ary_new2(tbl->num_entries);
st_foreach_safe(tbl, list_i, ary);
st_free_table(tbl);
return ary;
}
/*
* call-seq:
* mod.constants(inherit=true) -> array
*
* Returns an array of the names of the constants accessible in
* <i>mod</i>. This includes the names of constants in any included
* modules (example at start of section), unless the <i>inherit</i>
* parameter is set to <code>false</code>.
*
* The implementation makes no guarantees about the order in which the
* constants are yielded.
*
* IO.constants.include?(:SYNC) #=> true
* IO.constants(false).include?(:SYNC) #=> false
*
* Also see Module#const_defined?.
*/
VALUE
rb_mod_constants(int argc, const VALUE *argv, VALUE mod)
{
2019-11-25 09:05:53 +03:00
bool inherit = true;
if (rb_check_arity(argc, 0, 1)) inherit = RTEST(argv[0]);
if (inherit) {
return rb_const_list(rb_mod_const_of(mod, 0));
}
else {
return rb_local_constants(mod);
}
}
static int
rb_const_defined_0(VALUE klass, ID id, int exclude, int recurse, int visibility)
{
VALUE tmp;
int mod_retry = 0;
rb_const_entry_t *ce;
tmp = klass;
retry:
while (tmp) {
if ((ce = rb_const_lookup(tmp, id))) {
if (visibility && RB_CONST_PRIVATE_P(ce)) {
return (int)Qfalse;
}
if (ce->value == Qundef && !check_autoload_required(tmp, id, 0) &&
!rb_autoloading_value(tmp, id, NULL, NULL))
return (int)Qfalse;
if (exclude && tmp == rb_cObject && klass != rb_cObject) {
return (int)Qfalse;
}
return (int)Qtrue;
}
if (!recurse) break;
tmp = RCLASS_SUPER(tmp);
}
if (!exclude && !mod_retry && BUILTIN_TYPE(klass) == T_MODULE) {
mod_retry = 1;
tmp = rb_cObject;
goto retry;
}
return (int)Qfalse;
}
int
rb_const_defined_from(VALUE klass, ID id)
{
return rb_const_defined_0(klass, id, TRUE, TRUE, FALSE);
}
int
rb_const_defined(VALUE klass, ID id)
{
return rb_const_defined_0(klass, id, FALSE, TRUE, FALSE);
}
int
rb_const_defined_at(VALUE klass, ID id)
{
return rb_const_defined_0(klass, id, TRUE, FALSE, FALSE);
}
mjit_compile.c: merge initial JIT compiler which has been developed by Takashi Kokubun <takashikkbn@gmail> as YARV-MJIT. Many of its bugs are fixed by wanabe <s.wanabe@gmail.com>. This JIT compiler is designed to be a safe migration path to introduce JIT compiler to MRI. So this commit does not include any bytecode changes or dynamic instruction modifications, which are done in original MJIT. This commit even strips off some aggressive optimizations from YARV-MJIT, and thus it's slower than YARV-MJIT too. But it's still fairly faster than Ruby 2.5 in some benchmarks (attached below). Note that this JIT compiler passes `make test`, `make test-all`, `make test-spec` without JIT, and even with JIT. Not only it's perfectly safe with JIT disabled because it does not replace VM instructions unlike MJIT, but also with JIT enabled it stably runs Ruby applications including Rails applications. I'm expecting this version as just "initial" JIT compiler. I have many optimization ideas which are skipped for initial merging, and you may easily replace this JIT compiler with a faster one by just replacing mjit_compile.c. `mjit_compile` interface is designed for the purpose. common.mk: update dependencies for mjit_compile.c. internal.h: declare `rb_vm_insn_addr2insn` for MJIT. vm.c: exclude some definitions if `-DMJIT_HEADER` is provided to compiler. This avoids to include some functions which take a long time to compile, e.g. vm_exec_core. Some of the purpose is achieved in transform_mjit_header.rb (see `IGNORED_FUNCTIONS`) but others are manually resolved for now. Load mjit_helper.h for MJIT header. mjit_helper.h: New. This is a file used only by JIT-ed code. I'll refactor `mjit_call_cfunc` later. vm_eval.c: add some #ifdef switches to skip compiling some functions like Init_vm_eval. win32/mkexports.rb: export thread/ec functions, which are used by MJIT. include/ruby/defines.h: add MJIT_FUNC_EXPORTED macro alis to clarify that a function is exported only for MJIT. array.c: export a function used by MJIT. bignum.c: ditto. class.c: ditto. compile.c: ditto. error.c: ditto. gc.c: ditto. hash.c: ditto. iseq.c: ditto. numeric.c: ditto. object.c: ditto. proc.c: ditto. re.c: ditto. st.c: ditto. string.c: ditto. thread.c: ditto. variable.c: ditto. vm_backtrace.c: ditto. vm_insnhelper.c: ditto. vm_method.c: ditto. I would like to improve maintainability of function exports, but I believe this way is acceptable as initial merging if we clarify the new exports are for MJIT (so that we can use them as TODO list to fix) and add unit tests to detect unresolved symbols. I'll add unit tests of JIT compilations in succeeding commits. Author: Takashi Kokubun <takashikkbn@gmail.com> Contributor: wanabe <s.wanabe@gmail.com> Part of [Feature #14235] --- * Known issues * Code generated by gcc is faster than clang. The benchmark may be worse in macOS. Following benchmark result is provided by gcc w/ Linux. * Performance is decreased when Google Chrome is running * JIT can work on MinGW, but it doesn't improve performance at least in short running benchmark. * Currently it doesn't perform well with Rails. We'll try to fix this before release. --- * Benchmark reslts Benchmarked with: Intel 4.0GHz i7-4790K with 16GB memory under x86-64 Ubuntu 8 Cores - 2.0.0-p0: Ruby 2.0.0-p0 - r62186: Ruby trunk (early 2.6.0), before MJIT changes - JIT off: On this commit, but without `--jit` option - JIT on: On this commit, and with `--jit` option ** Optcarrot fps Benchmark: https://github.com/mame/optcarrot | |2.0.0-p0 |r62186 |JIT off |JIT on | |:--------|:--------|:--------|:--------|:--------| |fps |37.32 |51.46 |51.31 |58.88 | |vs 2.0.0 |1.00x |1.38x |1.37x |1.58x | ** MJIT benchmarks Benchmark: https://github.com/benchmark-driver/mjit-benchmarks (Original: https://github.com/vnmakarov/ruby/tree/rtl_mjit_branch/MJIT-benchmarks) | |2.0.0-p0 |r62186 |JIT off |JIT on | |:----------|:--------|:--------|:--------|:--------| |aread |1.00 |1.09 |1.07 |2.19 | |aref |1.00 |1.13 |1.11 |2.22 | |aset |1.00 |1.50 |1.45 |2.64 | |awrite |1.00 |1.17 |1.13 |2.20 | |call |1.00 |1.29 |1.26 |2.02 | |const2 |1.00 |1.10 |1.10 |2.19 | |const |1.00 |1.11 |1.10 |2.19 | |fannk |1.00 |1.04 |1.02 |1.00 | |fib |1.00 |1.32 |1.31 |1.84 | |ivread |1.00 |1.13 |1.12 |2.43 | |ivwrite |1.00 |1.23 |1.21 |2.40 | |mandelbrot |1.00 |1.13 |1.16 |1.28 | |meteor |1.00 |2.97 |2.92 |3.17 | |nbody |1.00 |1.17 |1.15 |1.49 | |nest-ntimes|1.00 |1.22 |1.20 |1.39 | |nest-while |1.00 |1.10 |1.10 |1.37 | |norm |1.00 |1.18 |1.16 |1.24 | |nsvb |1.00 |1.16 |1.16 |1.17 | |red-black |1.00 |1.02 |0.99 |1.12 | |sieve |1.00 |1.30 |1.28 |1.62 | |trees |1.00 |1.14 |1.13 |1.19 | |while |1.00 |1.12 |1.11 |2.41 | ** Discourse's script/bench.rb Benchmark: https://github.com/discourse/discourse/blob/v1.8.7/script/bench.rb NOTE: Rails performance was somehow a little degraded with JIT for now. We should fix this. (At least I know opt_aref is performing badly in JIT and I have an idea to fix it. Please wait for the fix.) *** JIT off Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 17 75: 18 90: 22 99: 29 home_admin: 50: 21 75: 21 90: 27 99: 40 topic_admin: 50: 17 75: 18 90: 22 99: 32 categories: 50: 35 75: 41 90: 43 99: 77 home: 50: 39 75: 46 90: 49 99: 95 topic: 50: 46 75: 52 90: 56 99: 101 *** JIT on Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 19 75: 21 90: 25 99: 33 home_admin: 50: 24 75: 26 90: 30 99: 35 topic_admin: 50: 19 75: 20 90: 25 99: 30 categories: 50: 40 75: 44 90: 48 99: 76 home: 50: 42 75: 48 90: 51 99: 89 topic: 50: 49 75: 55 90: 58 99: 99 git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62197 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-02-04 14:22:28 +03:00
MJIT_FUNC_EXPORTED int
rb_public_const_defined_from(VALUE klass, ID id)
{
return rb_const_defined_0(klass, id, TRUE, TRUE, TRUE);
}
static void
check_before_mod_set(VALUE klass, ID id, VALUE val, const char *dest)
{
rb_check_frozen(klass);
}
static void set_namespace_path(VALUE named_namespace, VALUE name);
static enum rb_id_table_iterator_result
set_namespace_path_i(ID id, VALUE v, void *payload)
{
rb_const_entry_t *ce = (rb_const_entry_t *)v;
VALUE value = ce->value;
int has_permanent_classpath;
VALUE parental_path = *((VALUE *) payload);
2021-10-06 14:18:35 +03:00
if (!rb_is_const_id(id) || !rb_namespace_p(value)) {
return ID_TABLE_CONTINUE;
}
classname(value, &has_permanent_classpath);
if (has_permanent_classpath) {
return ID_TABLE_CONTINUE;
}
set_namespace_path(value, build_const_path(parental_path, id));
if (RCLASS_IV_TBL(value)) {
st_data_t tmp = tmp_classpath;
st_delete(RCLASS_IV_TBL(value), &tmp, 0);
}
return ID_TABLE_CONTINUE;
}
/*
* Assign permanent classpaths to all namespaces that are directly or indirectly
* nested under +named_namespace+. +named_namespace+ must have a permanent
* classpath.
*/
static void
set_namespace_path(VALUE named_namespace, VALUE namespace_path)
{
struct rb_id_table *const_table = RCLASS_CONST_TBL(named_namespace);
RB_VM_LOCK_ENTER();
{
rb_class_ivar_set(named_namespace, classpath, namespace_path);
if (const_table) {
rb_id_table_foreach(const_table, set_namespace_path_i, &namespace_path);
}
}
RB_VM_LOCK_LEAVE();
}
void
rb_const_set(VALUE klass, ID id, VALUE val)
{
rb_const_entry_t *ce;
* sprintf.c (rb_str_format): allow %c to print one character string (e.g. ?x). * lib/tempfile.rb (Tempfile::make_tmpname): put dot between basename and pid. [ruby-talk:196272] * parse.y (do_block): remove -> style block. * parse.y (parser_yylex): remove tLAMBDA_ARG. * eval.c (rb_call0): binding for the return event hook should have consistent scope. [ruby-core:07928] * eval.c (proc_invoke): return behavior should depend whether it is surrounded by a lambda or a mere block. * eval.c (formal_assign): handles post splat arguments. * eval.c (rb_call0): ditto. * st.c (strhash): use FNV-1a hash. * parse.y (parser_yylex): removed experimental ';;' terminator. * eval.c (rb_node_arity): should be aware of post splat arguments. * eval.c (rb_proc_arity): ditto. * parse.y (f_args): syntax rule enhanced to support arguments after the splat. * parse.y (block_param): ditto for block parameters. * parse.y (f_post_arg): mandatory formal arguments after the splat argument. * parse.y (new_args_gen): generate nodes for mandatory formal arguments after the splat argument. * eval.c (rb_eval): dispatch mandatory formal arguments after the splat argument. * parse.y (args): allow more than one splat in the argument list. * parse.y (method_call): allow aref [] to accept all kind of method argument, including assocs, splat, and block argument. * eval.c (SETUP_ARGS0): prepare block argument as well. * lib/mathn.rb (Integer): remove Integer#gcd2. [ruby-core:07931] * eval.c (error_line): print receivers true/false/nil specially. * eval.c (rb_proc_yield): handles parameters in yield semantics. * eval.c (nil_yield): gives LocalJumpError to denote no block error. * io.c (rb_io_getc): now takes one-character string. * string.c (rb_str_hash): use FNV-1a hash from Fowler/Noll/Vo hashing algorithm. * string.c (rb_str_aref): str[0] now returns 1 character string, instead of a fixnum. [Ruby2] * parse.y (parser_yylex): ?c now returns 1 character string, instead of a fixnum. [Ruby2] * string.c (rb_str_aset): no longer support fixnum insertion. * eval.c (umethod_bind): should not update original class. [ruby-dev:28636] * eval.c (ev_const_get): should support constant access from within instance_eval(). [ruby-dev:28327] * time.c (time_timeval): should round for usec floating number. [ruby-core:07896] * time.c (time_add): ditto. * dir.c (sys_warning): should not call a vararg function rb_sys_warning() indirectly. [ruby-core:07886] * numeric.c (flo_divmod): the first element of Float#divmod should be an integer. [ruby-dev:28589] * test/ruby/test_float.rb: add tests for divmod, div, modulo and remainder. * re.c (rb_reg_initialize): should not allow modifying literal regexps. frozen check moved from rb_reg_initialize_m as well. * re.c (rb_reg_initialize): should not modify untainted objects in safe levels higher than 3. * re.c (rb_memcmp): type change from char* to const void*. * dir.c (dir_close): should not close untainted dir stream. * dir.c (GetDIR): add tainted/frozen check for each dir operation. * lib/rdoc/parsers/parse_rb.rb (RDoc::RubyParser::parse_symbol_arg): typo fixed. a patch from Florian Gross <florg at florg.net>. * eval.c (EXEC_EVENT_HOOK): trace_func may remove itself from event_hooks. no guarantee for arbitrary hook deletion. [ruby-dev:28632] * util.c (ruby_strtod): differ addition to minimize error. [ruby-dev:28619] * util.c (ruby_strtod): should not raise ERANGE when the input string does not have any digits. [ruby-dev:28629] * eval.c (proc_invoke): should restore old ruby_frame->block. thanks to ts <decoux at moulon.inra.fr>. [ruby-core:07833] also fix [ruby-dev:28614] as well. * signal.c (trap): sig should be less then NSIG. Coverity found this bug. a patch from Kevin Tew <tewk at tewk.com>. [ruby-core:07823] * math.c (math_log2): add new method inspired by [ruby-talk:191237]. * math.c (math_log): add optional base argument to Math::log(). [ruby-talk:191308] * ext/syck/emitter.c (syck_scan_scalar): avoid accessing uninitialized array element. a patch from Pat Eyler <rubypate at gmail.com>. [ruby-core:07809] * array.c (rb_ary_fill): initialize local variables first. a patch from Pat Eyler <rubypate at gmail.com>. [ruby-core:07810] * ext/syck/yaml2byte.c (syck_yaml2byte_handler): need to free type_tag. a patch from Pat Eyler <rubypate at gmail.com>. [ruby-core:07808] * ext/socket/socket.c (make_hostent_internal): accept ai_family check from Sam Roberts <sroberts at uniserve.com>. [ruby-core:07691] * util.c (ruby_strtod): should not cut off 18 digits for no reason. [ruby-core:07796] * array.c (rb_ary_fill): internalize local variable "beg" to pacify Coverity. [ruby-core:07770] * pack.c (pack_unpack): now supports CRLF newlines. a patch from <tommy at tmtm.org>. [ruby-dev:28601] * applied code clean-up patch from Stefan Huehner <stefan at huehner.org>. [ruby-core:07764] * lib/jcode.rb (String::tr_s): should have translated non squeezing character sequence (i.e. a character) as well. thanks to Hiroshi Ichikawa <gimite at gimite.ddo.jp> [ruby-list:42090] * ext/socket/socket.c: document update patch from Sam Roberts <sroberts at uniserve.com>. [ruby-core:07701] * lib/mathn.rb (Integer): need not to remove gcd2. a patch from NARUSE, Yui <naruse at airemix.com>. [ruby-dev:28570] * parse.y (arg): too much NEW_LIST() * eval.c (SETUP_ARGS0): remove unnecessary access to nd_alen. * eval.c (rb_eval): use ARGSCAT for NODE_OP_ASGN1. [ruby-dev:28585] * parse.y (arg): use NODE_ARGSCAT for placeholder. * lib/getoptlong.rb (GetoptLong::get): RDoc update patch from mathew <meta at pobox.com>. [ruby-core:07738] * variable.c (rb_const_set): raise error when no target klass is supplied. [ruby-dev:28582] * prec.c (prec_prec_f): documentation patch from <gerardo.santana at gmail.com>. [ruby-core:07689] * bignum.c (rb_big_pow): second operand may be too big even if it's a Fixnum. [ruby-talk:187984] * README.EXT: update symbol description. [ruby-talk:188104] * COPYING: explicitly note GPLv2. [ruby-talk:187922] * parse.y: remove some obsolete syntax rules (unparenthesized method calls in argument list). * eval.c (rb_call0): insecure calling should be checked for non NODE_SCOPE method invocations too. * eval.c (rb_alias): should preserve the current safe level as well as method definition. * process.c (rb_f_sleep): remove RDoc description about SIGALRM which is not valid on the current implementation. [ruby-dev:28464] Thu Mar 23 21:40:47 2006 K.Kosako <sndgk393 AT ybb.ne.jp> * eval.c (method_missing): should support argument splat in super. a bug in combination of super, splat and method_missing. [ruby-talk:185438] * configure.in: Solaris SunPro compiler -rapth patch from <kuwa at labs.fujitsu.com>. [ruby-dev:28443] * configure.in: remove enable_rpath=no for Solaris. [ruby-dev:28440] * ext/win32ole/win32ole.c (ole_val2olevariantdata): change behavior of converting OLE Variant object with VT_ARRAY|VT_UI1 and Ruby String object. * ruby.1: a clarification patch from David Lutterkort <dlutter at redhat.com>. [ruby-core:7508] * lib/rdoc/ri/ri_paths.rb (RI::Paths): adding paths from rubygems directories. a patch from Eric Hodel <drbrain at segment7.net>. [ruby-core:07423] * eval.c (rb_clear_cache_by_class): clearing wrong cache. * ext/extmk.rb: use :remove_destination to install extension libraries to avoid SEGV. [ruby-dev:28417] * eval.c (rb_thread_fd_writable): should not re-schedule output from KILLED thread (must be error printing). * array.c (rb_ary_flatten_bang): allow specifying recursion level. [ruby-talk:182170] * array.c (rb_ary_flatten): ditto. * gc.c (add_heap): a heap_slots may overflow. a patch from Stefan Weil <weil at mail.berlios.de>. * eval.c (rb_call): use separate cache for fcall/vcall invocation. * eval.c (rb_eval): NODE_FCALL, NODE_VCALL can call local functions. * eval.c (rb_mod_local): a new method to specify newly added visibility "local". * eval.c (search_method): search for local methods which are visible only from the current class. * class.c (rb_class_local_methods): a method to list local methods. * object.c (Init_Object): add BasicObject class as a top level BlankSlate class. * ruby.h (SYM2ID): should not cast to signed long. [ruby-core:07414] * class.c (rb_include_module): allow module duplication. git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@10235 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2006-06-10 01:20:17 +04:00
if (NIL_P(klass)) {
rb_raise(rb_eTypeError, "no class/module to define constant %"PRIsVALUE"",
QUOTE_ID(id));
* sprintf.c (rb_str_format): allow %c to print one character string (e.g. ?x). * lib/tempfile.rb (Tempfile::make_tmpname): put dot between basename and pid. [ruby-talk:196272] * parse.y (do_block): remove -> style block. * parse.y (parser_yylex): remove tLAMBDA_ARG. * eval.c (rb_call0): binding for the return event hook should have consistent scope. [ruby-core:07928] * eval.c (proc_invoke): return behavior should depend whether it is surrounded by a lambda or a mere block. * eval.c (formal_assign): handles post splat arguments. * eval.c (rb_call0): ditto. * st.c (strhash): use FNV-1a hash. * parse.y (parser_yylex): removed experimental ';;' terminator. * eval.c (rb_node_arity): should be aware of post splat arguments. * eval.c (rb_proc_arity): ditto. * parse.y (f_args): syntax rule enhanced to support arguments after the splat. * parse.y (block_param): ditto for block parameters. * parse.y (f_post_arg): mandatory formal arguments after the splat argument. * parse.y (new_args_gen): generate nodes for mandatory formal arguments after the splat argument. * eval.c (rb_eval): dispatch mandatory formal arguments after the splat argument. * parse.y (args): allow more than one splat in the argument list. * parse.y (method_call): allow aref [] to accept all kind of method argument, including assocs, splat, and block argument. * eval.c (SETUP_ARGS0): prepare block argument as well. * lib/mathn.rb (Integer): remove Integer#gcd2. [ruby-core:07931] * eval.c (error_line): print receivers true/false/nil specially. * eval.c (rb_proc_yield): handles parameters in yield semantics. * eval.c (nil_yield): gives LocalJumpError to denote no block error. * io.c (rb_io_getc): now takes one-character string. * string.c (rb_str_hash): use FNV-1a hash from Fowler/Noll/Vo hashing algorithm. * string.c (rb_str_aref): str[0] now returns 1 character string, instead of a fixnum. [Ruby2] * parse.y (parser_yylex): ?c now returns 1 character string, instead of a fixnum. [Ruby2] * string.c (rb_str_aset): no longer support fixnum insertion. * eval.c (umethod_bind): should not update original class. [ruby-dev:28636] * eval.c (ev_const_get): should support constant access from within instance_eval(). [ruby-dev:28327] * time.c (time_timeval): should round for usec floating number. [ruby-core:07896] * time.c (time_add): ditto. * dir.c (sys_warning): should not call a vararg function rb_sys_warning() indirectly. [ruby-core:07886] * numeric.c (flo_divmod): the first element of Float#divmod should be an integer. [ruby-dev:28589] * test/ruby/test_float.rb: add tests for divmod, div, modulo and remainder. * re.c (rb_reg_initialize): should not allow modifying literal regexps. frozen check moved from rb_reg_initialize_m as well. * re.c (rb_reg_initialize): should not modify untainted objects in safe levels higher than 3. * re.c (rb_memcmp): type change from char* to const void*. * dir.c (dir_close): should not close untainted dir stream. * dir.c (GetDIR): add tainted/frozen check for each dir operation. * lib/rdoc/parsers/parse_rb.rb (RDoc::RubyParser::parse_symbol_arg): typo fixed. a patch from Florian Gross <florg at florg.net>. * eval.c (EXEC_EVENT_HOOK): trace_func may remove itself from event_hooks. no guarantee for arbitrary hook deletion. [ruby-dev:28632] * util.c (ruby_strtod): differ addition to minimize error. [ruby-dev:28619] * util.c (ruby_strtod): should not raise ERANGE when the input string does not have any digits. [ruby-dev:28629] * eval.c (proc_invoke): should restore old ruby_frame->block. thanks to ts <decoux at moulon.inra.fr>. [ruby-core:07833] also fix [ruby-dev:28614] as well. * signal.c (trap): sig should be less then NSIG. Coverity found this bug. a patch from Kevin Tew <tewk at tewk.com>. [ruby-core:07823] * math.c (math_log2): add new method inspired by [ruby-talk:191237]. * math.c (math_log): add optional base argument to Math::log(). [ruby-talk:191308] * ext/syck/emitter.c (syck_scan_scalar): avoid accessing uninitialized array element. a patch from Pat Eyler <rubypate at gmail.com>. [ruby-core:07809] * array.c (rb_ary_fill): initialize local variables first. a patch from Pat Eyler <rubypate at gmail.com>. [ruby-core:07810] * ext/syck/yaml2byte.c (syck_yaml2byte_handler): need to free type_tag. a patch from Pat Eyler <rubypate at gmail.com>. [ruby-core:07808] * ext/socket/socket.c (make_hostent_internal): accept ai_family check from Sam Roberts <sroberts at uniserve.com>. [ruby-core:07691] * util.c (ruby_strtod): should not cut off 18 digits for no reason. [ruby-core:07796] * array.c (rb_ary_fill): internalize local variable "beg" to pacify Coverity. [ruby-core:07770] * pack.c (pack_unpack): now supports CRLF newlines. a patch from <tommy at tmtm.org>. [ruby-dev:28601] * applied code clean-up patch from Stefan Huehner <stefan at huehner.org>. [ruby-core:07764] * lib/jcode.rb (String::tr_s): should have translated non squeezing character sequence (i.e. a character) as well. thanks to Hiroshi Ichikawa <gimite at gimite.ddo.jp> [ruby-list:42090] * ext/socket/socket.c: document update patch from Sam Roberts <sroberts at uniserve.com>. [ruby-core:07701] * lib/mathn.rb (Integer): need not to remove gcd2. a patch from NARUSE, Yui <naruse at airemix.com>. [ruby-dev:28570] * parse.y (arg): too much NEW_LIST() * eval.c (SETUP_ARGS0): remove unnecessary access to nd_alen. * eval.c (rb_eval): use ARGSCAT for NODE_OP_ASGN1. [ruby-dev:28585] * parse.y (arg): use NODE_ARGSCAT for placeholder. * lib/getoptlong.rb (GetoptLong::get): RDoc update patch from mathew <meta at pobox.com>. [ruby-core:07738] * variable.c (rb_const_set): raise error when no target klass is supplied. [ruby-dev:28582] * prec.c (prec_prec_f): documentation patch from <gerardo.santana at gmail.com>. [ruby-core:07689] * bignum.c (rb_big_pow): second operand may be too big even if it's a Fixnum. [ruby-talk:187984] * README.EXT: update symbol description. [ruby-talk:188104] * COPYING: explicitly note GPLv2. [ruby-talk:187922] * parse.y: remove some obsolete syntax rules (unparenthesized method calls in argument list). * eval.c (rb_call0): insecure calling should be checked for non NODE_SCOPE method invocations too. * eval.c (rb_alias): should preserve the current safe level as well as method definition. * process.c (rb_f_sleep): remove RDoc description about SIGALRM which is not valid on the current implementation. [ruby-dev:28464] Thu Mar 23 21:40:47 2006 K.Kosako <sndgk393 AT ybb.ne.jp> * eval.c (method_missing): should support argument splat in super. a bug in combination of super, splat and method_missing. [ruby-talk:185438] * configure.in: Solaris SunPro compiler -rapth patch from <kuwa at labs.fujitsu.com>. [ruby-dev:28443] * configure.in: remove enable_rpath=no for Solaris. [ruby-dev:28440] * ext/win32ole/win32ole.c (ole_val2olevariantdata): change behavior of converting OLE Variant object with VT_ARRAY|VT_UI1 and Ruby String object. * ruby.1: a clarification patch from David Lutterkort <dlutter at redhat.com>. [ruby-core:7508] * lib/rdoc/ri/ri_paths.rb (RI::Paths): adding paths from rubygems directories. a patch from Eric Hodel <drbrain at segment7.net>. [ruby-core:07423] * eval.c (rb_clear_cache_by_class): clearing wrong cache. * ext/extmk.rb: use :remove_destination to install extension libraries to avoid SEGV. [ruby-dev:28417] * eval.c (rb_thread_fd_writable): should not re-schedule output from KILLED thread (must be error printing). * array.c (rb_ary_flatten_bang): allow specifying recursion level. [ruby-talk:182170] * array.c (rb_ary_flatten): ditto. * gc.c (add_heap): a heap_slots may overflow. a patch from Stefan Weil <weil at mail.berlios.de>. * eval.c (rb_call): use separate cache for fcall/vcall invocation. * eval.c (rb_eval): NODE_FCALL, NODE_VCALL can call local functions. * eval.c (rb_mod_local): a new method to specify newly added visibility "local". * eval.c (search_method): search for local methods which are visible only from the current class. * class.c (rb_class_local_methods): a method to list local methods. * object.c (Init_Object): add BasicObject class as a top level BlankSlate class. * ruby.h (SYM2ID): should not cast to signed long. [ruby-core:07414] * class.c (rb_include_module): allow module duplication. git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@10235 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2006-06-10 01:20:17 +04:00
}
if (!rb_ractor_main_p() && !rb_ractor_shareable_p(val)) {
rb_raise(rb_eRactorIsolationError, "can not set constants with non-shareable objects by non-main Ractors");
}
check_before_mod_set(klass, id, val, "constant");
RB_VM_LOCK_ENTER();
{
struct rb_id_table *tbl = RCLASS_CONST_TBL(klass);
if (!tbl) {
RCLASS_CONST_TBL(klass) = tbl = rb_id_table_create(0);
rb_clear_constant_cache();
ce = ZALLOC(rb_const_entry_t);
rb_id_table_insert(tbl, id, (VALUE)ce);
setup_const_entry(ce, klass, val, CONST_PUBLIC);
}
else {
struct autoload_const ac = {
.mod = klass, .id = id,
.value = val, .flag = CONST_PUBLIC,
/* fill the rest with 0 */
};
const_tbl_update(&ac);
}
}
RB_VM_LOCK_LEAVE();
/*
* Resolve and cache class name immediately to resolve ambiguity
* and avoid order-dependency on const_tbl
*/
if (rb_cObject && rb_namespace_p(val)) {
int val_path_permanent;
VALUE val_path = classname(val, &val_path_permanent);
if (NIL_P(val_path) || !val_path_permanent) {
if (klass == rb_cObject) {
set_namespace_path(val, rb_id2str(id));
}
else {
int parental_path_permanent;
VALUE parental_path = classname(klass, &parental_path_permanent);
if (NIL_P(parental_path)) {
int throwaway;
parental_path = rb_tmp_class_path(klass, &throwaway, make_temporary_path);
}
if (parental_path_permanent && !val_path_permanent) {
set_namespace_path(val, build_const_path(parental_path, id));
}
else if (!parental_path_permanent && NIL_P(val_path)) {
ivar_set(val, tmp_classpath, build_const_path(parental_path, id));
}
}
}
}
}
static struct autoload_data_i *
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
current_autoload_data(VALUE mod, ID id, struct autoload_const **acp)
{
struct autoload_data_i *ele;
VALUE load = autoload_data(mod, id);
if (!load) return 0;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
ele = get_autoload_data(load, acp);
if (!ele) return 0;
/* for autoloading thread, keep the defined value to autoloading storage */
if (ele->state && (ele->state->thread == rb_thread_current())) {
return ele;
}
return 0;
}
static void
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
const_tbl_update(struct autoload_const *ac)
{
VALUE value;
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
VALUE klass = ac->mod;
VALUE val = ac->value;
ID id = ac->id;
struct rb_id_table *tbl = RCLASS_CONST_TBL(klass);
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
rb_const_flag_t visibility = ac->flag;
rb_const_entry_t *ce;
if (rb_id_table_lookup(tbl, id, &value)) {
ce = (rb_const_entry_t *)value;
if (ce->value == Qundef) {
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
struct autoload_data_i *ele = current_autoload_data(klass, id, &ac);
if (ele) {
rb_clear_constant_cache();
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
ac->value = val; /* autoload_i is non-WB-protected */
ac->file = rb_source_location(&ac->line);
}
else {
/* otherwise autoloaded constant, allow to override */
autoload_delete(klass, id);
ce->flag = visibility;
RB_OBJ_WRITE(klass, &ce->value, val);
RB_OBJ_WRITE(klass, &ce->file, ac->file);
ce->line = ac->line;
}
return;
}
else {
VALUE name = QUOTE_ID(id);
visibility = ce->flag;
if (klass == rb_cObject)
rb_warn("already initialized constant %"PRIsVALUE"", name);
else
rb_warn("already initialized constant %"PRIsVALUE"::%"PRIsVALUE"",
rb_class_name(klass), name);
if (!NIL_P(ce->file) && ce->line) {
rb_compile_warn(RSTRING_PTR(ce->file), ce->line,
"previous definition of %"PRIsVALUE" was here", name);
}
}
rb_clear_constant_cache();
setup_const_entry(ce, klass, val, visibility);
}
else {
rb_clear_constant_cache();
ce = ZALLOC(rb_const_entry_t);
rb_id_table_insert(tbl, id, (VALUE)ce);
setup_const_entry(ce, klass, val, visibility);
}
}
static void
setup_const_entry(rb_const_entry_t *ce, VALUE klass, VALUE val,
rb_const_flag_t visibility)
{
ce->flag = visibility;
RB_OBJ_WRITE(klass, &ce->value, val);
RB_OBJ_WRITE(klass, &ce->file, rb_source_location(&ce->line));
}
void
rb_define_const(VALUE klass, const char *name, VALUE val)
{
ID id = rb_intern(name);
if (!rb_is_const_id(id)) {
rb_warn("rb_define_const: invalid name `%s' for constant", name);
}
rb_gc_register_mark_object(val);
rb_const_set(klass, id, val);
}
void
rb_define_global_const(const char *name, VALUE val)
{
rb_define_const(rb_cObject, name, val);
}
static void
set_const_visibility(VALUE mod, int argc, const VALUE *argv,
rb_const_flag_t flag, rb_const_flag_t mask)
{
int i;
rb_const_entry_t *ce;
ID id;
rb_class_modify_check(mod);
if (argc == 0) {
rb_warning("%"PRIsVALUE" with no argument is just ignored",
QUOTE_ID(rb_frame_callee()));
return;
}
for (i = 0; i < argc; i++) {
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
struct autoload_const *ac;
VALUE val = argv[i];
id = rb_check_id(&val);
if (!id) {
if (i > 0) {
rb_clear_constant_cache();
}
undefined_constant(mod, val);
}
if ((ce = rb_const_lookup(mod, id))) {
ce->flag &= ~mask;
ce->flag |= flag;
if (ce->value == Qundef) {
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
struct autoload_data_i *ele;
ele = current_autoload_data(mod, id, &ac);
if (ele) {
variable.c: fix multiple autoload with identical file (again) We need to ensure autoload declarations pointing to the same feature (aka "file") can wait on each other to avoid deadlock situations. So, reorganize autoload data structures to maintain a feature => autoload_data_i mapping, and have module constant tables point to the new autoload_const struct instead of directly to autoload_data_i. This allows multiple autoload_const structs to refer to the SAME autoload_data_i struct, and with it, the on-stack autoload_state.waitq. The end result is different constants can share the same waitq (tied to the feature name), and not deadlock each other during loading. Thanks to Eugene Kenny for the bug report and reproducible test case. Reported-by: Eugene Kenny <elkenny@gmail.com> * variable.c (autoload_featuremap): new global (struct autoload_const): new per-const struct (struct autoload_state): reference autoload_const instead of autoload_data_i (struct autoload_data_i): remove per-const (autoload_i_mark): delete from autoload_featuremap if unreferenced (autoload_c_mark): new dmark callback (autoload_c_free): new dfree callback (autoload_c_memsize): new memsize callback (autoload_const_type): new data type (get_autoload_data): set autoload_const as well (rb_autoload_str): use new data structures (autoload_delete): cleanup from autoload_featuremap (check_autoload_required): adjust for new internals (rb_autoloading_value): ditto (struct autoload_const_set_args): remove, redundant with autoload_const (const_tbl_update): adjust for new internals (autoload_const_set): ditto (autoload_require): ditto (autoload_reset): ditto (rb_autoload_load): ditto (rb_const_set): ditto (current_autoload_data): ditto (set_const_visibility): ditto * test/ruby/test_autoload.rb (test_autoload_same_file): new test (test_no_leak): new test [ruby-core:86935] [Bug #14742] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63392 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-05-10 11:54:26 +03:00
ac->flag &= ~mask;
ac->flag |= flag;
}
}
}
else {
if (i > 0) {
rb_clear_constant_cache();
}
undefined_constant(mod, ID2SYM(id));
}
}
rb_clear_constant_cache();
}
void
rb_deprecate_constant(VALUE mod, const char *name)
{
rb_const_entry_t *ce;
ID id;
long len = strlen(name);
rb_class_modify_check(mod);
if (!(id = rb_check_id_cstr(name, len, NULL))) {
undefined_constant(mod, rb_fstring_new(name, len));
}
if (!(ce = rb_const_lookup(mod, id))) {
undefined_constant(mod, ID2SYM(id));
}
ce->flag |= CONST_DEPRECATED;
}
/*
* call-seq:
* mod.private_constant(symbol, ...) => mod
*
* Makes a list of existing constants private.
*/
VALUE
rb_mod_private_constant(int argc, const VALUE *argv, VALUE obj)
{
set_const_visibility(obj, argc, argv, CONST_PRIVATE, CONST_VISIBILITY_MASK);
return obj;
}
/*
* call-seq:
* mod.public_constant(symbol, ...) => mod
*
* Makes a list of existing constants public.
*/
VALUE
rb_mod_public_constant(int argc, const VALUE *argv, VALUE obj)
{
set_const_visibility(obj, argc, argv, CONST_PUBLIC, CONST_VISIBILITY_MASK);
return obj;
}
/*
* call-seq:
* mod.deprecate_constant(symbol, ...) => mod
*
* Makes a list of existing constants deprecated. Attempt
* to refer to them will produce a warning.
*
* module HTTP
* NotFound = Exception.new
* NOT_FOUND = NotFound # previous version of the library used this name
*
* deprecate_constant :NOT_FOUND
* end
*
* HTTP::NOT_FOUND
* # warning: constant HTTP::NOT_FOUND is deprecated
*
*/
VALUE
rb_mod_deprecate_constant(int argc, const VALUE *argv, VALUE obj)
{
set_const_visibility(obj, argc, argv, CONST_DEPRECATED, CONST_DEPRECATED);
return obj;
}
static VALUE
original_module(VALUE c)
{
if (RB_TYPE_P(c, T_ICLASS))
return RBASIC(c)->klass;
return c;
}
static int
cvar_lookup_at(VALUE klass, ID id, st_data_t *v)
{
if (!RCLASS_IV_TBL(klass)) return 0;
return st_lookup(RCLASS_IV_TBL(klass), (st_data_t)id, v);
}
static VALUE
cvar_front_klass(VALUE klass)
{
if (FL_TEST(klass, FL_SINGLETON)) {
VALUE obj = rb_ivar_get(klass, id__attached__);
if (rb_namespace_p(obj)) {
return obj;
}
}
return RCLASS_SUPER(klass);
}
static void
cvar_overtaken(VALUE front, VALUE target, ID id)
{
if (front && target != front) {
st_data_t did = (st_data_t)id;
if (original_module(front) != original_module(target)) {
rb_raise(rb_eRuntimeError,
"class variable % "PRIsVALUE" of %"PRIsVALUE" is overtaken by %"PRIsVALUE"",
ID2SYM(id), rb_class_name(original_module(front)),
rb_class_name(original_module(target)));
}
if (BUILTIN_TYPE(front) == T_CLASS) {
st_delete(RCLASS_IV_TBL(front), &did, 0);
}
}
}
Add a cache for class variables Redo of 34a2acdac788602c14bf05fb616215187badd504 and 931138b00696419945dc03e10f033b1f53cd50f3 which were reverted. GitHub PR #4340. This change implements a cache for class variables. Previously there was no cache for cvars. Cvar access is slow due to needing to travel all the way up th ancestor tree before returning the cvar value. The deeper the ancestor tree the slower cvar access will be. The benefits of the cache are more visible with a higher number of included modules due to the way Ruby looks up class variables. The benchmark here includes 26 modules and shows with the cache, this branch is 6.5x faster when accessing class variables. ``` compare-ruby: ruby 3.1.0dev (2021-03-15T06:22:34Z master 9e5105c) [x86_64-darwin19] built-ruby: ruby 3.1.0dev (2021-03-15T12:12:44Z add-cache-for-clas.. c6be009) [x86_64-darwin19] | |compare-ruby|built-ruby| |:--------|-----------:|---------:| |vm_cvar | 5.681M| 36.980M| | | -| 6.51x| ``` Benchmark.ips calling `ActiveRecord::Base.logger` from within a Rails application. ActiveRecord::Base.logger has 71 ancestors. The more ancestors a tree has, the more clear the speed increase. IE if Base had only one ancestor we'd see no improvement. This benchmark is run on a vanilla Rails application. Benchmark code: ```ruby require "benchmark/ips" require_relative "config/environment" Benchmark.ips do |x| x.report "logger" do ActiveRecord::Base.logger end end ``` Ruby 3.0 master / Rails 6.1: ``` Warming up -------------------------------------- logger 155.251k i/100ms Calculating ------------------------------------- ``` Ruby 3.0 with cvar cache / Rails 6.1: ``` Warming up -------------------------------------- logger 1.546M i/100ms Calculating ------------------------------------- logger 14.857M (± 4.8%) i/s - 74.198M in 5.006202s ``` Lastly we ran a benchmark to demonstate the difference between master and our cache when the number of modules increases. This benchmark measures 1 ancestor, 30 ancestors, and 100 ancestors. Ruby 3.0 master: ``` Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower ``` Ruby 3.0 with cvar cache: ``` Warming up -------------------------------------- 1 module 1.641M i/100ms 30 modules 1.655M i/100ms 100 modules 1.620M i/100ms Calculating ------------------------------------- 1 module 16.279M (± 3.8%) i/s - 82.038M in 5.046923s 30 modules 15.891M (± 3.9%) i/s - 79.459M in 5.007958s 100 modules 16.087M (± 3.6%) i/s - 81.005M in 5.041931s Comparison: 1 module: 16279458.0 i/s 100 modules: 16087484.6 i/s - same-ish: difference falls within error 30 modules: 15891406.2 i/s - same-ish: difference falls within error ``` Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
2021-06-01 20:34:06 +03:00
static VALUE
find_cvar(VALUE klass, VALUE * front, VALUE * target, ID id)
{
VALUE v = Qundef;
CVAR_ACCESSOR_SHOULD_BE_MAIN_RACTOR();
if (cvar_lookup_at(klass, id, (&v))) {
if (!*front) {
*front = klass;
}
*target = klass;
}
for (klass = cvar_front_klass(klass); klass; klass = RCLASS_SUPER(klass)) {
if (cvar_lookup_at(klass, id, (&v))) {
if (!*front) {
*front = klass;
}
*target = klass;
}
}
return v;
}
#define CVAR_FOREACH_ANCESTORS(klass, v, r) \
for (klass = cvar_front_klass(klass); klass; klass = RCLASS_SUPER(klass)) { \
if (cvar_lookup_at(klass, id, (v))) { \
r; \
} \
}
#define CVAR_LOOKUP(v,r) do {\
CVAR_ACCESSOR_SHOULD_BE_MAIN_RACTOR(); \
if (cvar_lookup_at(klass, id, (v))) {r;}\
CVAR_FOREACH_ANCESTORS(klass, v, r);\
} while(0)
Add a cache for class variables Redo of 34a2acdac788602c14bf05fb616215187badd504 and 931138b00696419945dc03e10f033b1f53cd50f3 which were reverted. GitHub PR #4340. This change implements a cache for class variables. Previously there was no cache for cvars. Cvar access is slow due to needing to travel all the way up th ancestor tree before returning the cvar value. The deeper the ancestor tree the slower cvar access will be. The benefits of the cache are more visible with a higher number of included modules due to the way Ruby looks up class variables. The benchmark here includes 26 modules and shows with the cache, this branch is 6.5x faster when accessing class variables. ``` compare-ruby: ruby 3.1.0dev (2021-03-15T06:22:34Z master 9e5105c) [x86_64-darwin19] built-ruby: ruby 3.1.0dev (2021-03-15T12:12:44Z add-cache-for-clas.. c6be009) [x86_64-darwin19] | |compare-ruby|built-ruby| |:--------|-----------:|---------:| |vm_cvar | 5.681M| 36.980M| | | -| 6.51x| ``` Benchmark.ips calling `ActiveRecord::Base.logger` from within a Rails application. ActiveRecord::Base.logger has 71 ancestors. The more ancestors a tree has, the more clear the speed increase. IE if Base had only one ancestor we'd see no improvement. This benchmark is run on a vanilla Rails application. Benchmark code: ```ruby require "benchmark/ips" require_relative "config/environment" Benchmark.ips do |x| x.report "logger" do ActiveRecord::Base.logger end end ``` Ruby 3.0 master / Rails 6.1: ``` Warming up -------------------------------------- logger 155.251k i/100ms Calculating ------------------------------------- ``` Ruby 3.0 with cvar cache / Rails 6.1: ``` Warming up -------------------------------------- logger 1.546M i/100ms Calculating ------------------------------------- logger 14.857M (± 4.8%) i/s - 74.198M in 5.006202s ``` Lastly we ran a benchmark to demonstate the difference between master and our cache when the number of modules increases. This benchmark measures 1 ancestor, 30 ancestors, and 100 ancestors. Ruby 3.0 master: ``` Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower ``` Ruby 3.0 with cvar cache: ``` Warming up -------------------------------------- 1 module 1.641M i/100ms 30 modules 1.655M i/100ms 100 modules 1.620M i/100ms Calculating ------------------------------------- 1 module 16.279M (± 3.8%) i/s - 82.038M in 5.046923s 30 modules 15.891M (± 3.9%) i/s - 79.459M in 5.007958s 100 modules 16.087M (± 3.6%) i/s - 81.005M in 5.041931s Comparison: 1 module: 16279458.0 i/s 100 modules: 16087484.6 i/s - same-ish: difference falls within error 30 modules: 15891406.2 i/s - same-ish: difference falls within error ``` Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
2021-06-01 20:34:06 +03:00
static void
check_for_cvar_table(VALUE subclass, VALUE key)
{
st_table *tbl = RCLASS_IV_TBL(subclass);
if (tbl && st_lookup(tbl, key, NULL)) {
RB_DEBUG_COUNTER_INC(cvar_class_invalidate);
ruby_vm_global_cvar_state++;
return;
}
rb_class_foreach_subclass(subclass, check_for_cvar_table, key);
}
void
rb_cvar_set(VALUE klass, ID id, VALUE val)
{
VALUE tmp, front = 0, target = 0;
tmp = klass;
CVAR_LOOKUP(0, {if (!front) front = klass; target = klass;});
if (target) {
cvar_overtaken(front, target, id);
}
else {
target = tmp;
}
if (RB_TYPE_P(target, T_ICLASS)) {
target = RBASIC(target)->klass;
}
check_before_mod_set(target, id, val, "class variable");
Add a cache for class variables Redo of 34a2acdac788602c14bf05fb616215187badd504 and 931138b00696419945dc03e10f033b1f53cd50f3 which were reverted. GitHub PR #4340. This change implements a cache for class variables. Previously there was no cache for cvars. Cvar access is slow due to needing to travel all the way up th ancestor tree before returning the cvar value. The deeper the ancestor tree the slower cvar access will be. The benefits of the cache are more visible with a higher number of included modules due to the way Ruby looks up class variables. The benchmark here includes 26 modules and shows with the cache, this branch is 6.5x faster when accessing class variables. ``` compare-ruby: ruby 3.1.0dev (2021-03-15T06:22:34Z master 9e5105c) [x86_64-darwin19] built-ruby: ruby 3.1.0dev (2021-03-15T12:12:44Z add-cache-for-clas.. c6be009) [x86_64-darwin19] | |compare-ruby|built-ruby| |:--------|-----------:|---------:| |vm_cvar | 5.681M| 36.980M| | | -| 6.51x| ``` Benchmark.ips calling `ActiveRecord::Base.logger` from within a Rails application. ActiveRecord::Base.logger has 71 ancestors. The more ancestors a tree has, the more clear the speed increase. IE if Base had only one ancestor we'd see no improvement. This benchmark is run on a vanilla Rails application. Benchmark code: ```ruby require "benchmark/ips" require_relative "config/environment" Benchmark.ips do |x| x.report "logger" do ActiveRecord::Base.logger end end ``` Ruby 3.0 master / Rails 6.1: ``` Warming up -------------------------------------- logger 155.251k i/100ms Calculating ------------------------------------- ``` Ruby 3.0 with cvar cache / Rails 6.1: ``` Warming up -------------------------------------- logger 1.546M i/100ms Calculating ------------------------------------- logger 14.857M (± 4.8%) i/s - 74.198M in 5.006202s ``` Lastly we ran a benchmark to demonstate the difference between master and our cache when the number of modules increases. This benchmark measures 1 ancestor, 30 ancestors, and 100 ancestors. Ruby 3.0 master: ``` Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower ``` Ruby 3.0 with cvar cache: ``` Warming up -------------------------------------- 1 module 1.641M i/100ms 30 modules 1.655M i/100ms 100 modules 1.620M i/100ms Calculating ------------------------------------- 1 module 16.279M (± 3.8%) i/s - 82.038M in 5.046923s 30 modules 15.891M (± 3.9%) i/s - 79.459M in 5.007958s 100 modules 16.087M (± 3.6%) i/s - 81.005M in 5.041931s Comparison: 1 module: 16279458.0 i/s 100 modules: 16087484.6 i/s - same-ish: difference falls within error 30 modules: 15891406.2 i/s - same-ish: difference falls within error ``` Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
2021-06-01 20:34:06 +03:00
int result = rb_class_ivar_set(target, id, val);
struct rb_id_table *rb_cvc_tbl = RCLASS_CVC_TBL(target);
if (!rb_cvc_tbl) {
rb_cvc_tbl = RCLASS_CVC_TBL(target) = rb_id_table_create(2);
}
struct rb_cvar_class_tbl_entry *ent;
VALUE ent_data;
Add a cache for class variables Redo of 34a2acdac788602c14bf05fb616215187badd504 and 931138b00696419945dc03e10f033b1f53cd50f3 which were reverted. GitHub PR #4340. This change implements a cache for class variables. Previously there was no cache for cvars. Cvar access is slow due to needing to travel all the way up th ancestor tree before returning the cvar value. The deeper the ancestor tree the slower cvar access will be. The benefits of the cache are more visible with a higher number of included modules due to the way Ruby looks up class variables. The benchmark here includes 26 modules and shows with the cache, this branch is 6.5x faster when accessing class variables. ``` compare-ruby: ruby 3.1.0dev (2021-03-15T06:22:34Z master 9e5105c) [x86_64-darwin19] built-ruby: ruby 3.1.0dev (2021-03-15T12:12:44Z add-cache-for-clas.. c6be009) [x86_64-darwin19] | |compare-ruby|built-ruby| |:--------|-----------:|---------:| |vm_cvar | 5.681M| 36.980M| | | -| 6.51x| ``` Benchmark.ips calling `ActiveRecord::Base.logger` from within a Rails application. ActiveRecord::Base.logger has 71 ancestors. The more ancestors a tree has, the more clear the speed increase. IE if Base had only one ancestor we'd see no improvement. This benchmark is run on a vanilla Rails application. Benchmark code: ```ruby require "benchmark/ips" require_relative "config/environment" Benchmark.ips do |x| x.report "logger" do ActiveRecord::Base.logger end end ``` Ruby 3.0 master / Rails 6.1: ``` Warming up -------------------------------------- logger 155.251k i/100ms Calculating ------------------------------------- ``` Ruby 3.0 with cvar cache / Rails 6.1: ``` Warming up -------------------------------------- logger 1.546M i/100ms Calculating ------------------------------------- logger 14.857M (± 4.8%) i/s - 74.198M in 5.006202s ``` Lastly we ran a benchmark to demonstate the difference between master and our cache when the number of modules increases. This benchmark measures 1 ancestor, 30 ancestors, and 100 ancestors. Ruby 3.0 master: ``` Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower ``` Ruby 3.0 with cvar cache: ``` Warming up -------------------------------------- 1 module 1.641M i/100ms 30 modules 1.655M i/100ms 100 modules 1.620M i/100ms Calculating ------------------------------------- 1 module 16.279M (± 3.8%) i/s - 82.038M in 5.046923s 30 modules 15.891M (± 3.9%) i/s - 79.459M in 5.007958s 100 modules 16.087M (± 3.6%) i/s - 81.005M in 5.041931s Comparison: 1 module: 16279458.0 i/s 100 modules: 16087484.6 i/s - same-ish: difference falls within error 30 modules: 15891406.2 i/s - same-ish: difference falls within error ``` Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
2021-06-01 20:34:06 +03:00
if (!rb_id_table_lookup(rb_cvc_tbl, id, &ent_data)) {
Add a cache for class variables Redo of 34a2acdac788602c14bf05fb616215187badd504 and 931138b00696419945dc03e10f033b1f53cd50f3 which were reverted. GitHub PR #4340. This change implements a cache for class variables. Previously there was no cache for cvars. Cvar access is slow due to needing to travel all the way up th ancestor tree before returning the cvar value. The deeper the ancestor tree the slower cvar access will be. The benefits of the cache are more visible with a higher number of included modules due to the way Ruby looks up class variables. The benchmark here includes 26 modules and shows with the cache, this branch is 6.5x faster when accessing class variables. ``` compare-ruby: ruby 3.1.0dev (2021-03-15T06:22:34Z master 9e5105c) [x86_64-darwin19] built-ruby: ruby 3.1.0dev (2021-03-15T12:12:44Z add-cache-for-clas.. c6be009) [x86_64-darwin19] | |compare-ruby|built-ruby| |:--------|-----------:|---------:| |vm_cvar | 5.681M| 36.980M| | | -| 6.51x| ``` Benchmark.ips calling `ActiveRecord::Base.logger` from within a Rails application. ActiveRecord::Base.logger has 71 ancestors. The more ancestors a tree has, the more clear the speed increase. IE if Base had only one ancestor we'd see no improvement. This benchmark is run on a vanilla Rails application. Benchmark code: ```ruby require "benchmark/ips" require_relative "config/environment" Benchmark.ips do |x| x.report "logger" do ActiveRecord::Base.logger end end ``` Ruby 3.0 master / Rails 6.1: ``` Warming up -------------------------------------- logger 155.251k i/100ms Calculating ------------------------------------- ``` Ruby 3.0 with cvar cache / Rails 6.1: ``` Warming up -------------------------------------- logger 1.546M i/100ms Calculating ------------------------------------- logger 14.857M (± 4.8%) i/s - 74.198M in 5.006202s ``` Lastly we ran a benchmark to demonstate the difference between master and our cache when the number of modules increases. This benchmark measures 1 ancestor, 30 ancestors, and 100 ancestors. Ruby 3.0 master: ``` Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower ``` Ruby 3.0 with cvar cache: ``` Warming up -------------------------------------- 1 module 1.641M i/100ms 30 modules 1.655M i/100ms 100 modules 1.620M i/100ms Calculating ------------------------------------- 1 module 16.279M (± 3.8%) i/s - 82.038M in 5.046923s 30 modules 15.891M (± 3.9%) i/s - 79.459M in 5.007958s 100 modules 16.087M (± 3.6%) i/s - 81.005M in 5.041931s Comparison: 1 module: 16279458.0 i/s 100 modules: 16087484.6 i/s - same-ish: difference falls within error 30 modules: 15891406.2 i/s - same-ish: difference falls within error ``` Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
2021-06-01 20:34:06 +03:00
ent = ALLOC(struct rb_cvar_class_tbl_entry);
ent->class_value = target;
ent->global_cvar_state = GET_GLOBAL_CVAR_STATE();
rb_id_table_insert(rb_cvc_tbl, id, (VALUE)ent);
RB_DEBUG_COUNTER_INC(cvar_inline_miss);
}
else {
ent = (void *)ent_data;
Add a cache for class variables Redo of 34a2acdac788602c14bf05fb616215187badd504 and 931138b00696419945dc03e10f033b1f53cd50f3 which were reverted. GitHub PR #4340. This change implements a cache for class variables. Previously there was no cache for cvars. Cvar access is slow due to needing to travel all the way up th ancestor tree before returning the cvar value. The deeper the ancestor tree the slower cvar access will be. The benefits of the cache are more visible with a higher number of included modules due to the way Ruby looks up class variables. The benchmark here includes 26 modules and shows with the cache, this branch is 6.5x faster when accessing class variables. ``` compare-ruby: ruby 3.1.0dev (2021-03-15T06:22:34Z master 9e5105c) [x86_64-darwin19] built-ruby: ruby 3.1.0dev (2021-03-15T12:12:44Z add-cache-for-clas.. c6be009) [x86_64-darwin19] | |compare-ruby|built-ruby| |:--------|-----------:|---------:| |vm_cvar | 5.681M| 36.980M| | | -| 6.51x| ``` Benchmark.ips calling `ActiveRecord::Base.logger` from within a Rails application. ActiveRecord::Base.logger has 71 ancestors. The more ancestors a tree has, the more clear the speed increase. IE if Base had only one ancestor we'd see no improvement. This benchmark is run on a vanilla Rails application. Benchmark code: ```ruby require "benchmark/ips" require_relative "config/environment" Benchmark.ips do |x| x.report "logger" do ActiveRecord::Base.logger end end ``` Ruby 3.0 master / Rails 6.1: ``` Warming up -------------------------------------- logger 155.251k i/100ms Calculating ------------------------------------- ``` Ruby 3.0 with cvar cache / Rails 6.1: ``` Warming up -------------------------------------- logger 1.546M i/100ms Calculating ------------------------------------- logger 14.857M (± 4.8%) i/s - 74.198M in 5.006202s ``` Lastly we ran a benchmark to demonstate the difference between master and our cache when the number of modules increases. This benchmark measures 1 ancestor, 30 ancestors, and 100 ancestors. Ruby 3.0 master: ``` Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower ``` Ruby 3.0 with cvar cache: ``` Warming up -------------------------------------- 1 module 1.641M i/100ms 30 modules 1.655M i/100ms 100 modules 1.620M i/100ms Calculating ------------------------------------- 1 module 16.279M (± 3.8%) i/s - 82.038M in 5.046923s 30 modules 15.891M (± 3.9%) i/s - 79.459M in 5.007958s 100 modules 16.087M (± 3.6%) i/s - 81.005M in 5.041931s Comparison: 1 module: 16279458.0 i/s 100 modules: 16087484.6 i/s - same-ish: difference falls within error 30 modules: 15891406.2 i/s - same-ish: difference falls within error ``` Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
2021-06-01 20:34:06 +03:00
ent->global_cvar_state = GET_GLOBAL_CVAR_STATE();
}
// Break the cvar cache if this is a new class variable
// and target is a module or a subclass with the same
// cvar in this lookup.
if (result == 0) {
if (RB_TYPE_P(target, T_CLASS)) {
if (RCLASS_SUBCLASSES(target)) {
rb_class_foreach_subclass(target, check_for_cvar_table, id);
}
}
}
}
VALUE
Add a cache for class variables Redo of 34a2acdac788602c14bf05fb616215187badd504 and 931138b00696419945dc03e10f033b1f53cd50f3 which were reverted. GitHub PR #4340. This change implements a cache for class variables. Previously there was no cache for cvars. Cvar access is slow due to needing to travel all the way up th ancestor tree before returning the cvar value. The deeper the ancestor tree the slower cvar access will be. The benefits of the cache are more visible with a higher number of included modules due to the way Ruby looks up class variables. The benchmark here includes 26 modules and shows with the cache, this branch is 6.5x faster when accessing class variables. ``` compare-ruby: ruby 3.1.0dev (2021-03-15T06:22:34Z master 9e5105c) [x86_64-darwin19] built-ruby: ruby 3.1.0dev (2021-03-15T12:12:44Z add-cache-for-clas.. c6be009) [x86_64-darwin19] | |compare-ruby|built-ruby| |:--------|-----------:|---------:| |vm_cvar | 5.681M| 36.980M| | | -| 6.51x| ``` Benchmark.ips calling `ActiveRecord::Base.logger` from within a Rails application. ActiveRecord::Base.logger has 71 ancestors. The more ancestors a tree has, the more clear the speed increase. IE if Base had only one ancestor we'd see no improvement. This benchmark is run on a vanilla Rails application. Benchmark code: ```ruby require "benchmark/ips" require_relative "config/environment" Benchmark.ips do |x| x.report "logger" do ActiveRecord::Base.logger end end ``` Ruby 3.0 master / Rails 6.1: ``` Warming up -------------------------------------- logger 155.251k i/100ms Calculating ------------------------------------- ``` Ruby 3.0 with cvar cache / Rails 6.1: ``` Warming up -------------------------------------- logger 1.546M i/100ms Calculating ------------------------------------- logger 14.857M (± 4.8%) i/s - 74.198M in 5.006202s ``` Lastly we ran a benchmark to demonstate the difference between master and our cache when the number of modules increases. This benchmark measures 1 ancestor, 30 ancestors, and 100 ancestors. Ruby 3.0 master: ``` Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower ``` Ruby 3.0 with cvar cache: ``` Warming up -------------------------------------- 1 module 1.641M i/100ms 30 modules 1.655M i/100ms 100 modules 1.620M i/100ms Calculating ------------------------------------- 1 module 16.279M (± 3.8%) i/s - 82.038M in 5.046923s 30 modules 15.891M (± 3.9%) i/s - 79.459M in 5.007958s 100 modules 16.087M (± 3.6%) i/s - 81.005M in 5.041931s Comparison: 1 module: 16279458.0 i/s 100 modules: 16087484.6 i/s - same-ish: difference falls within error 30 modules: 15891406.2 i/s - same-ish: difference falls within error ``` Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
2021-06-01 20:34:06 +03:00
rb_cvar_find(VALUE klass, ID id, VALUE *front)
{
Add a cache for class variables Redo of 34a2acdac788602c14bf05fb616215187badd504 and 931138b00696419945dc03e10f033b1f53cd50f3 which were reverted. GitHub PR #4340. This change implements a cache for class variables. Previously there was no cache for cvars. Cvar access is slow due to needing to travel all the way up th ancestor tree before returning the cvar value. The deeper the ancestor tree the slower cvar access will be. The benefits of the cache are more visible with a higher number of included modules due to the way Ruby looks up class variables. The benchmark here includes 26 modules and shows with the cache, this branch is 6.5x faster when accessing class variables. ``` compare-ruby: ruby 3.1.0dev (2021-03-15T06:22:34Z master 9e5105c) [x86_64-darwin19] built-ruby: ruby 3.1.0dev (2021-03-15T12:12:44Z add-cache-for-clas.. c6be009) [x86_64-darwin19] | |compare-ruby|built-ruby| |:--------|-----------:|---------:| |vm_cvar | 5.681M| 36.980M| | | -| 6.51x| ``` Benchmark.ips calling `ActiveRecord::Base.logger` from within a Rails application. ActiveRecord::Base.logger has 71 ancestors. The more ancestors a tree has, the more clear the speed increase. IE if Base had only one ancestor we'd see no improvement. This benchmark is run on a vanilla Rails application. Benchmark code: ```ruby require "benchmark/ips" require_relative "config/environment" Benchmark.ips do |x| x.report "logger" do ActiveRecord::Base.logger end end ``` Ruby 3.0 master / Rails 6.1: ``` Warming up -------------------------------------- logger 155.251k i/100ms Calculating ------------------------------------- ``` Ruby 3.0 with cvar cache / Rails 6.1: ``` Warming up -------------------------------------- logger 1.546M i/100ms Calculating ------------------------------------- logger 14.857M (± 4.8%) i/s - 74.198M in 5.006202s ``` Lastly we ran a benchmark to demonstate the difference between master and our cache when the number of modules increases. This benchmark measures 1 ancestor, 30 ancestors, and 100 ancestors. Ruby 3.0 master: ``` Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower ``` Ruby 3.0 with cvar cache: ``` Warming up -------------------------------------- 1 module 1.641M i/100ms 30 modules 1.655M i/100ms 100 modules 1.620M i/100ms Calculating ------------------------------------- 1 module 16.279M (± 3.8%) i/s - 82.038M in 5.046923s 30 modules 15.891M (± 3.9%) i/s - 79.459M in 5.007958s 100 modules 16.087M (± 3.6%) i/s - 81.005M in 5.041931s Comparison: 1 module: 16279458.0 i/s 100 modules: 16087484.6 i/s - same-ish: difference falls within error 30 modules: 15891406.2 i/s - same-ish: difference falls within error ``` Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
2021-06-01 20:34:06 +03:00
VALUE target = 0;
VALUE value;
Add a cache for class variables Redo of 34a2acdac788602c14bf05fb616215187badd504 and 931138b00696419945dc03e10f033b1f53cd50f3 which were reverted. GitHub PR #4340. This change implements a cache for class variables. Previously there was no cache for cvars. Cvar access is slow due to needing to travel all the way up th ancestor tree before returning the cvar value. The deeper the ancestor tree the slower cvar access will be. The benefits of the cache are more visible with a higher number of included modules due to the way Ruby looks up class variables. The benchmark here includes 26 modules and shows with the cache, this branch is 6.5x faster when accessing class variables. ``` compare-ruby: ruby 3.1.0dev (2021-03-15T06:22:34Z master 9e5105c) [x86_64-darwin19] built-ruby: ruby 3.1.0dev (2021-03-15T12:12:44Z add-cache-for-clas.. c6be009) [x86_64-darwin19] | |compare-ruby|built-ruby| |:--------|-----------:|---------:| |vm_cvar | 5.681M| 36.980M| | | -| 6.51x| ``` Benchmark.ips calling `ActiveRecord::Base.logger` from within a Rails application. ActiveRecord::Base.logger has 71 ancestors. The more ancestors a tree has, the more clear the speed increase. IE if Base had only one ancestor we'd see no improvement. This benchmark is run on a vanilla Rails application. Benchmark code: ```ruby require "benchmark/ips" require_relative "config/environment" Benchmark.ips do |x| x.report "logger" do ActiveRecord::Base.logger end end ``` Ruby 3.0 master / Rails 6.1: ``` Warming up -------------------------------------- logger 155.251k i/100ms Calculating ------------------------------------- ``` Ruby 3.0 with cvar cache / Rails 6.1: ``` Warming up -------------------------------------- logger 1.546M i/100ms Calculating ------------------------------------- logger 14.857M (± 4.8%) i/s - 74.198M in 5.006202s ``` Lastly we ran a benchmark to demonstate the difference between master and our cache when the number of modules increases. This benchmark measures 1 ancestor, 30 ancestors, and 100 ancestors. Ruby 3.0 master: ``` Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower ``` Ruby 3.0 with cvar cache: ``` Warming up -------------------------------------- 1 module 1.641M i/100ms 30 modules 1.655M i/100ms 100 modules 1.620M i/100ms Calculating ------------------------------------- 1 module 16.279M (± 3.8%) i/s - 82.038M in 5.046923s 30 modules 15.891M (± 3.9%) i/s - 79.459M in 5.007958s 100 modules 16.087M (± 3.6%) i/s - 81.005M in 5.041931s Comparison: 1 module: 16279458.0 i/s 100 modules: 16087484.6 i/s - same-ish: difference falls within error 30 modules: 15891406.2 i/s - same-ish: difference falls within error ``` Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
2021-06-01 20:34:06 +03:00
value = find_cvar(klass, front, &target, id);
if (!target) {
rb_name_err_raise("uninitialized class variable %1$s in %2$s",
Add a cache for class variables Redo of 34a2acdac788602c14bf05fb616215187badd504 and 931138b00696419945dc03e10f033b1f53cd50f3 which were reverted. GitHub PR #4340. This change implements a cache for class variables. Previously there was no cache for cvars. Cvar access is slow due to needing to travel all the way up th ancestor tree before returning the cvar value. The deeper the ancestor tree the slower cvar access will be. The benefits of the cache are more visible with a higher number of included modules due to the way Ruby looks up class variables. The benchmark here includes 26 modules and shows with the cache, this branch is 6.5x faster when accessing class variables. ``` compare-ruby: ruby 3.1.0dev (2021-03-15T06:22:34Z master 9e5105c) [x86_64-darwin19] built-ruby: ruby 3.1.0dev (2021-03-15T12:12:44Z add-cache-for-clas.. c6be009) [x86_64-darwin19] | |compare-ruby|built-ruby| |:--------|-----------:|---------:| |vm_cvar | 5.681M| 36.980M| | | -| 6.51x| ``` Benchmark.ips calling `ActiveRecord::Base.logger` from within a Rails application. ActiveRecord::Base.logger has 71 ancestors. The more ancestors a tree has, the more clear the speed increase. IE if Base had only one ancestor we'd see no improvement. This benchmark is run on a vanilla Rails application. Benchmark code: ```ruby require "benchmark/ips" require_relative "config/environment" Benchmark.ips do |x| x.report "logger" do ActiveRecord::Base.logger end end ``` Ruby 3.0 master / Rails 6.1: ``` Warming up -------------------------------------- logger 155.251k i/100ms Calculating ------------------------------------- ``` Ruby 3.0 with cvar cache / Rails 6.1: ``` Warming up -------------------------------------- logger 1.546M i/100ms Calculating ------------------------------------- logger 14.857M (± 4.8%) i/s - 74.198M in 5.006202s ``` Lastly we ran a benchmark to demonstate the difference between master and our cache when the number of modules increases. This benchmark measures 1 ancestor, 30 ancestors, and 100 ancestors. Ruby 3.0 master: ``` Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower ``` Ruby 3.0 with cvar cache: ``` Warming up -------------------------------------- 1 module 1.641M i/100ms 30 modules 1.655M i/100ms 100 modules 1.620M i/100ms Calculating ------------------------------------- 1 module 16.279M (± 3.8%) i/s - 82.038M in 5.046923s 30 modules 15.891M (± 3.9%) i/s - 79.459M in 5.007958s 100 modules 16.087M (± 3.6%) i/s - 81.005M in 5.041931s Comparison: 1 module: 16279458.0 i/s 100 modules: 16087484.6 i/s - same-ish: difference falls within error 30 modules: 15891406.2 i/s - same-ish: difference falls within error ``` Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
2021-06-01 20:34:06 +03:00
klass, ID2SYM(id));
}
Add a cache for class variables Redo of 34a2acdac788602c14bf05fb616215187badd504 and 931138b00696419945dc03e10f033b1f53cd50f3 which were reverted. GitHub PR #4340. This change implements a cache for class variables. Previously there was no cache for cvars. Cvar access is slow due to needing to travel all the way up th ancestor tree before returning the cvar value. The deeper the ancestor tree the slower cvar access will be. The benefits of the cache are more visible with a higher number of included modules due to the way Ruby looks up class variables. The benchmark here includes 26 modules and shows with the cache, this branch is 6.5x faster when accessing class variables. ``` compare-ruby: ruby 3.1.0dev (2021-03-15T06:22:34Z master 9e5105c) [x86_64-darwin19] built-ruby: ruby 3.1.0dev (2021-03-15T12:12:44Z add-cache-for-clas.. c6be009) [x86_64-darwin19] | |compare-ruby|built-ruby| |:--------|-----------:|---------:| |vm_cvar | 5.681M| 36.980M| | | -| 6.51x| ``` Benchmark.ips calling `ActiveRecord::Base.logger` from within a Rails application. ActiveRecord::Base.logger has 71 ancestors. The more ancestors a tree has, the more clear the speed increase. IE if Base had only one ancestor we'd see no improvement. This benchmark is run on a vanilla Rails application. Benchmark code: ```ruby require "benchmark/ips" require_relative "config/environment" Benchmark.ips do |x| x.report "logger" do ActiveRecord::Base.logger end end ``` Ruby 3.0 master / Rails 6.1: ``` Warming up -------------------------------------- logger 155.251k i/100ms Calculating ------------------------------------- ``` Ruby 3.0 with cvar cache / Rails 6.1: ``` Warming up -------------------------------------- logger 1.546M i/100ms Calculating ------------------------------------- logger 14.857M (± 4.8%) i/s - 74.198M in 5.006202s ``` Lastly we ran a benchmark to demonstate the difference between master and our cache when the number of modules increases. This benchmark measures 1 ancestor, 30 ancestors, and 100 ancestors. Ruby 3.0 master: ``` Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower ``` Ruby 3.0 with cvar cache: ``` Warming up -------------------------------------- 1 module 1.641M i/100ms 30 modules 1.655M i/100ms 100 modules 1.620M i/100ms Calculating ------------------------------------- 1 module 16.279M (± 3.8%) i/s - 82.038M in 5.046923s 30 modules 15.891M (± 3.9%) i/s - 79.459M in 5.007958s 100 modules 16.087M (± 3.6%) i/s - 81.005M in 5.041931s Comparison: 1 module: 16279458.0 i/s 100 modules: 16087484.6 i/s - same-ish: difference falls within error 30 modules: 15891406.2 i/s - same-ish: difference falls within error ``` Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
2021-06-01 20:34:06 +03:00
cvar_overtaken(*front, target, id);
return (VALUE)value;
}
Add a cache for class variables Redo of 34a2acdac788602c14bf05fb616215187badd504 and 931138b00696419945dc03e10f033b1f53cd50f3 which were reverted. GitHub PR #4340. This change implements a cache for class variables. Previously there was no cache for cvars. Cvar access is slow due to needing to travel all the way up th ancestor tree before returning the cvar value. The deeper the ancestor tree the slower cvar access will be. The benefits of the cache are more visible with a higher number of included modules due to the way Ruby looks up class variables. The benchmark here includes 26 modules and shows with the cache, this branch is 6.5x faster when accessing class variables. ``` compare-ruby: ruby 3.1.0dev (2021-03-15T06:22:34Z master 9e5105c) [x86_64-darwin19] built-ruby: ruby 3.1.0dev (2021-03-15T12:12:44Z add-cache-for-clas.. c6be009) [x86_64-darwin19] | |compare-ruby|built-ruby| |:--------|-----------:|---------:| |vm_cvar | 5.681M| 36.980M| | | -| 6.51x| ``` Benchmark.ips calling `ActiveRecord::Base.logger` from within a Rails application. ActiveRecord::Base.logger has 71 ancestors. The more ancestors a tree has, the more clear the speed increase. IE if Base had only one ancestor we'd see no improvement. This benchmark is run on a vanilla Rails application. Benchmark code: ```ruby require "benchmark/ips" require_relative "config/environment" Benchmark.ips do |x| x.report "logger" do ActiveRecord::Base.logger end end ``` Ruby 3.0 master / Rails 6.1: ``` Warming up -------------------------------------- logger 155.251k i/100ms Calculating ------------------------------------- ``` Ruby 3.0 with cvar cache / Rails 6.1: ``` Warming up -------------------------------------- logger 1.546M i/100ms Calculating ------------------------------------- logger 14.857M (± 4.8%) i/s - 74.198M in 5.006202s ``` Lastly we ran a benchmark to demonstate the difference between master and our cache when the number of modules increases. This benchmark measures 1 ancestor, 30 ancestors, and 100 ancestors. Ruby 3.0 master: ``` Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower ``` Ruby 3.0 with cvar cache: ``` Warming up -------------------------------------- 1 module 1.641M i/100ms 30 modules 1.655M i/100ms 100 modules 1.620M i/100ms Calculating ------------------------------------- 1 module 16.279M (± 3.8%) i/s - 82.038M in 5.046923s 30 modules 15.891M (± 3.9%) i/s - 79.459M in 5.007958s 100 modules 16.087M (± 3.6%) i/s - 81.005M in 5.041931s Comparison: 1 module: 16279458.0 i/s 100 modules: 16087484.6 i/s - same-ish: difference falls within error 30 modules: 15891406.2 i/s - same-ish: difference falls within error ``` Co-authored-by: Aaron Patterson <tenderlove@ruby-lang.org>
2021-06-01 20:34:06 +03:00
VALUE
rb_cvar_get(VALUE klass, ID id)
{
VALUE front = 0;
return rb_cvar_find(klass, id, &front);
}
VALUE
rb_cvar_defined(VALUE klass, ID id)
{
if (!klass) return Qfalse;
CVAR_LOOKUP(0,return Qtrue);
return Qfalse;
}
static ID
cv_intern(VALUE klass, const char *name)
{
* eval.c (block_pass): should not downgrade safe level. * ext/dbm/extconf.rb: allow specifying dbm-type explicitly. * ext/dbm/extconf.rb: avoid gdbm if possible, because it leaks memory, whereas gdbm.so doesn't. potential incompatibility. * string.c (rb_str_insert): new method. * parse.y (yylex): lex_state after RESCUE_MOD should be EXPR_BEG. * array.c (rb_ary_insert): new method. * array.c (rb_ary_update): new utility function. * io.c (set_outfile): should check if closed before assignment. * eval.c (rb_eval): should preserve value of ruby_errinfo. * eval.c (rb_thread_schedule): infinite sleep should not cause dead lock. * array.c (rb_ary_flatten_bang): proper recursive detection. * eval.c (yield_under): need not to prohibit at safe level 4. * pack.c (pack_pack): p/P packs nil into NULL. * pack.c (pack_unpack): p/P unpacks NULL into nil. * pack.c (pack_pack): size check for P template. * ruby.c (set_arg0): wrong predicate when new $0 value is bigger than original space. * gc.c (id2ref): should use NUM2ULONG() * object.c (rb_mod_const_get): check whether name is a class variable name. * object.c (rb_mod_const_set): ditto. * object.c (rb_mod_const_defined): ditto. * marshal.c (w_float): precision changed to "%.16g" * eval.c (rb_call0): wrong retry behavior. * numeric.c (fix_aref): a bug on long>int architecture. * eval.c (rb_eval_string_wrap): should restore ruby_wrapper. * regex.c (re_compile_pattern): char class at either edge of range should be invalid. * eval.c (handle_rescue): use === to compare exception match. * error.c (syserr_eqq): comparison between SytemCallErrors should based on their error numbers. * eval.c (safe_getter): should use INT2NUM(). * bignum.c (rb_big2long): 2**31 cannot fit in 31 bit long. * regex.c (calculate_must_string): wrong length calculation. * eval.c (rb_thread_start_0): fixed memory leak. * parse.y (none): should clear cmdarg_stack too. * io.c (rb_fopen): use setvbuf() to avoid recursive malloc() on some platforms. * file.c (rb_stat_dev): device functions should honor stat field types (except long long such as dev_t). * eval.c (rb_mod_nesting): should not push nil for nesting array. * eval.c (rb_mod_s_constants): should not search array by rb_mod_const_at() for nil (happens for singleton class). * class.c (rb_singleton_class_attached): should modify iv_tbl by itself, no longer use rb_iv_set() to avoid freeze check error. * variable.c (rb_const_get): error message "uninitialized constant Foo at Bar::Baz" instead of "uninitialized constantBar::Baz::Foo". * eval.c (rb_mod_included): new hook called from rb_mod_include(). * io.c (opt_i_set): should strdup() inplace_edit string. * eval.c (exec_under): need to push cref too. * eval.c (rb_f_missing): raise NameError for "undefined local variable or method". * error.c (Init_Exception): new exception NoMethodError. NameError moved under ScriptError again. * eval.c (rb_f_missing): use NoMethodError instead of NameError. * file.c (Init_File): should redifine "new" class method. * eval.c (PUSH_CREF): sharing cref node was problematic. maintain runtime cref list instead. * eval.c (rb_eval): copy defn node before registering. * eval.c (rb_load): clear ruby_cref before loading. * variable.c (rb_const_get): no recursion to show full class path for modules. * eval.c (rb_set_safe_level): should set safe level in curr_thread as well. * eval.c (safe_setter): ditto. * object.c (rb_obj_is_instance_of): nil belongs to false, not true. * time.c (make_time_t): proper (I hope) daylight saving time handling for both US and Europe. I HATE DST! * eval.c (rb_thread_wait_for): non blocked signal interrupt should stop the interval. * eval.c (proc_eq): class check aded. * eval.c (proc_eq): typo fixed ("return" was ommitted). * error.c (Init_Exception): move NameError under StandardError. * class.c (rb_mod_clone): should copy method bodies too. * bignum.c (bigdivrem): should trim trailing zero bdigits of remainder, even if dd == 0. * file.c (check3rdbyte): safe string check moved here. * time.c (make_time_t): remove HAVE_TM_ZONE code since it sometimes reports wrong time. * time.c (make_time_t): remove unnecessary range check for platforms where negative time_t is available. * process.c (proc_waitall): should push Process::Status instead of Finuxm status. * process.c (waitall_each): should add all entries in pid_tbl. these changes are inspired by Koji Arai. Thanks. * process.c (proc_wait): should not iterate if pid_tbl is 0. * process.c (proc_waitall): ditto. * numeric.c (flodivmod): a bug in no fmod case. * process.c (pst_wifsignaled): should apply WIFSIGNALED for status (int), not st (VALUE). * io.c (Init_IO): value of $/ and $\ are no longer restricted to strings. type checks are done on demand. * class.c (rb_include_module): module inclusion should be check taints. * ruby.h (STR2CSTR): replace to StringType() and StringTypePtr(). * ruby.h (rb_str2cstr): ditto. * eval.c (rb_load): should not copy topleve local variables. It cause variable/method ambiguity. Thanks to L. Peter Deutsch. * class.c (rb_include_module): freeze check at first. * eval.c (rb_attr): sprintf() and rb_intern() moved into conditional body. git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@1356 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2001-05-02 08:22:21 +04:00
ID id = rb_intern(name);
if (!rb_is_class_id(id)) {
rb_name_err_raise("wrong class variable name %1$s",
klass, rb_str_new_cstr(name));
* eval.c (block_pass): should not downgrade safe level. * ext/dbm/extconf.rb: allow specifying dbm-type explicitly. * ext/dbm/extconf.rb: avoid gdbm if possible, because it leaks memory, whereas gdbm.so doesn't. potential incompatibility. * string.c (rb_str_insert): new method. * parse.y (yylex): lex_state after RESCUE_MOD should be EXPR_BEG. * array.c (rb_ary_insert): new method. * array.c (rb_ary_update): new utility function. * io.c (set_outfile): should check if closed before assignment. * eval.c (rb_eval): should preserve value of ruby_errinfo. * eval.c (rb_thread_schedule): infinite sleep should not cause dead lock. * array.c (rb_ary_flatten_bang): proper recursive detection. * eval.c (yield_under): need not to prohibit at safe level 4. * pack.c (pack_pack): p/P packs nil into NULL. * pack.c (pack_unpack): p/P unpacks NULL into nil. * pack.c (pack_pack): size check for P template. * ruby.c (set_arg0): wrong predicate when new $0 value is bigger than original space. * gc.c (id2ref): should use NUM2ULONG() * object.c (rb_mod_const_get): check whether name is a class variable name. * object.c (rb_mod_const_set): ditto. * object.c (rb_mod_const_defined): ditto. * marshal.c (w_float): precision changed to "%.16g" * eval.c (rb_call0): wrong retry behavior. * numeric.c (fix_aref): a bug on long>int architecture. * eval.c (rb_eval_string_wrap): should restore ruby_wrapper. * regex.c (re_compile_pattern): char class at either edge of range should be invalid. * eval.c (handle_rescue): use === to compare exception match. * error.c (syserr_eqq): comparison between SytemCallErrors should based on their error numbers. * eval.c (safe_getter): should use INT2NUM(). * bignum.c (rb_big2long): 2**31 cannot fit in 31 bit long. * regex.c (calculate_must_string): wrong length calculation. * eval.c (rb_thread_start_0): fixed memory leak. * parse.y (none): should clear cmdarg_stack too. * io.c (rb_fopen): use setvbuf() to avoid recursive malloc() on some platforms. * file.c (rb_stat_dev): device functions should honor stat field types (except long long such as dev_t). * eval.c (rb_mod_nesting): should not push nil for nesting array. * eval.c (rb_mod_s_constants): should not search array by rb_mod_const_at() for nil (happens for singleton class). * class.c (rb_singleton_class_attached): should modify iv_tbl by itself, no longer use rb_iv_set() to avoid freeze check error. * variable.c (rb_const_get): error message "uninitialized constant Foo at Bar::Baz" instead of "uninitialized constantBar::Baz::Foo". * eval.c (rb_mod_included): new hook called from rb_mod_include(). * io.c (opt_i_set): should strdup() inplace_edit string. * eval.c (exec_under): need to push cref too. * eval.c (rb_f_missing): raise NameError for "undefined local variable or method". * error.c (Init_Exception): new exception NoMethodError. NameError moved under ScriptError again. * eval.c (rb_f_missing): use NoMethodError instead of NameError. * file.c (Init_File): should redifine "new" class method. * eval.c (PUSH_CREF): sharing cref node was problematic. maintain runtime cref list instead. * eval.c (rb_eval): copy defn node before registering. * eval.c (rb_load): clear ruby_cref before loading. * variable.c (rb_const_get): no recursion to show full class path for modules. * eval.c (rb_set_safe_level): should set safe level in curr_thread as well. * eval.c (safe_setter): ditto. * object.c (rb_obj_is_instance_of): nil belongs to false, not true. * time.c (make_time_t): proper (I hope) daylight saving time handling for both US and Europe. I HATE DST! * eval.c (rb_thread_wait_for): non blocked signal interrupt should stop the interval. * eval.c (proc_eq): class check aded. * eval.c (proc_eq): typo fixed ("return" was ommitted). * error.c (Init_Exception): move NameError under StandardError. * class.c (rb_mod_clone): should copy method bodies too. * bignum.c (bigdivrem): should trim trailing zero bdigits of remainder, even if dd == 0. * file.c (check3rdbyte): safe string check moved here. * time.c (make_time_t): remove HAVE_TM_ZONE code since it sometimes reports wrong time. * time.c (make_time_t): remove unnecessary range check for platforms where negative time_t is available. * process.c (proc_waitall): should push Process::Status instead of Finuxm status. * process.c (waitall_each): should add all entries in pid_tbl. these changes are inspired by Koji Arai. Thanks. * process.c (proc_wait): should not iterate if pid_tbl is 0. * process.c (proc_waitall): ditto. * numeric.c (flodivmod): a bug in no fmod case. * process.c (pst_wifsignaled): should apply WIFSIGNALED for status (int), not st (VALUE). * io.c (Init_IO): value of $/ and $\ are no longer restricted to strings. type checks are done on demand. * class.c (rb_include_module): module inclusion should be check taints. * ruby.h (STR2CSTR): replace to StringType() and StringTypePtr(). * ruby.h (rb_str2cstr): ditto. * eval.c (rb_load): should not copy topleve local variables. It cause variable/method ambiguity. Thanks to L. Peter Deutsch. * class.c (rb_include_module): freeze check at first. * eval.c (rb_attr): sprintf() and rb_intern() moved into conditional body. git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@1356 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2001-05-02 08:22:21 +04:00
}
return id;
}
void
rb_cv_set(VALUE klass, const char *name, VALUE val)
{
ID id = cv_intern(klass, name);
rb_cvar_set(klass, id, val);
}
VALUE
rb_cv_get(VALUE klass, const char *name)
{
ID id = cv_intern(klass, name);
* eval.c (block_pass): should not downgrade safe level. * ext/dbm/extconf.rb: allow specifying dbm-type explicitly. * ext/dbm/extconf.rb: avoid gdbm if possible, because it leaks memory, whereas gdbm.so doesn't. potential incompatibility. * string.c (rb_str_insert): new method. * parse.y (yylex): lex_state after RESCUE_MOD should be EXPR_BEG. * array.c (rb_ary_insert): new method. * array.c (rb_ary_update): new utility function. * io.c (set_outfile): should check if closed before assignment. * eval.c (rb_eval): should preserve value of ruby_errinfo. * eval.c (rb_thread_schedule): infinite sleep should not cause dead lock. * array.c (rb_ary_flatten_bang): proper recursive detection. * eval.c (yield_under): need not to prohibit at safe level 4. * pack.c (pack_pack): p/P packs nil into NULL. * pack.c (pack_unpack): p/P unpacks NULL into nil. * pack.c (pack_pack): size check for P template. * ruby.c (set_arg0): wrong predicate when new $0 value is bigger than original space. * gc.c (id2ref): should use NUM2ULONG() * object.c (rb_mod_const_get): check whether name is a class variable name. * object.c (rb_mod_const_set): ditto. * object.c (rb_mod_const_defined): ditto. * marshal.c (w_float): precision changed to "%.16g" * eval.c (rb_call0): wrong retry behavior. * numeric.c (fix_aref): a bug on long>int architecture. * eval.c (rb_eval_string_wrap): should restore ruby_wrapper. * regex.c (re_compile_pattern): char class at either edge of range should be invalid. * eval.c (handle_rescue): use === to compare exception match. * error.c (syserr_eqq): comparison between SytemCallErrors should based on their error numbers. * eval.c (safe_getter): should use INT2NUM(). * bignum.c (rb_big2long): 2**31 cannot fit in 31 bit long. * regex.c (calculate_must_string): wrong length calculation. * eval.c (rb_thread_start_0): fixed memory leak. * parse.y (none): should clear cmdarg_stack too. * io.c (rb_fopen): use setvbuf() to avoid recursive malloc() on some platforms. * file.c (rb_stat_dev): device functions should honor stat field types (except long long such as dev_t). * eval.c (rb_mod_nesting): should not push nil for nesting array. * eval.c (rb_mod_s_constants): should not search array by rb_mod_const_at() for nil (happens for singleton class). * class.c (rb_singleton_class_attached): should modify iv_tbl by itself, no longer use rb_iv_set() to avoid freeze check error. * variable.c (rb_const_get): error message "uninitialized constant Foo at Bar::Baz" instead of "uninitialized constantBar::Baz::Foo". * eval.c (rb_mod_included): new hook called from rb_mod_include(). * io.c (opt_i_set): should strdup() inplace_edit string. * eval.c (exec_under): need to push cref too. * eval.c (rb_f_missing): raise NameError for "undefined local variable or method". * error.c (Init_Exception): new exception NoMethodError. NameError moved under ScriptError again. * eval.c (rb_f_missing): use NoMethodError instead of NameError. * file.c (Init_File): should redifine "new" class method. * eval.c (PUSH_CREF): sharing cref node was problematic. maintain runtime cref list instead. * eval.c (rb_eval): copy defn node before registering. * eval.c (rb_load): clear ruby_cref before loading. * variable.c (rb_const_get): no recursion to show full class path for modules. * eval.c (rb_set_safe_level): should set safe level in curr_thread as well. * eval.c (safe_setter): ditto. * object.c (rb_obj_is_instance_of): nil belongs to false, not true. * time.c (make_time_t): proper (I hope) daylight saving time handling for both US and Europe. I HATE DST! * eval.c (rb_thread_wait_for): non blocked signal interrupt should stop the interval. * eval.c (proc_eq): class check aded. * eval.c (proc_eq): typo fixed ("return" was ommitted). * error.c (Init_Exception): move NameError under StandardError. * class.c (rb_mod_clone): should copy method bodies too. * bignum.c (bigdivrem): should trim trailing zero bdigits of remainder, even if dd == 0. * file.c (check3rdbyte): safe string check moved here. * time.c (make_time_t): remove HAVE_TM_ZONE code since it sometimes reports wrong time. * time.c (make_time_t): remove unnecessary range check for platforms where negative time_t is available. * process.c (proc_waitall): should push Process::Status instead of Finuxm status. * process.c (waitall_each): should add all entries in pid_tbl. these changes are inspired by Koji Arai. Thanks. * process.c (proc_wait): should not iterate if pid_tbl is 0. * process.c (proc_waitall): ditto. * numeric.c (flodivmod): a bug in no fmod case. * process.c (pst_wifsignaled): should apply WIFSIGNALED for status (int), not st (VALUE). * io.c (Init_IO): value of $/ and $\ are no longer restricted to strings. type checks are done on demand. * class.c (rb_include_module): module inclusion should be check taints. * ruby.h (STR2CSTR): replace to StringType() and StringTypePtr(). * ruby.h (rb_str2cstr): ditto. * eval.c (rb_load): should not copy topleve local variables. It cause variable/method ambiguity. Thanks to L. Peter Deutsch. * class.c (rb_include_module): freeze check at first. * eval.c (rb_attr): sprintf() and rb_intern() moved into conditional body. git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@1356 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2001-05-02 08:22:21 +04:00
return rb_cvar_get(klass, id);
}
void
rb_define_class_variable(VALUE klass, const char *name, VALUE val)
{
rb_cv_set(klass, name, val);
}
static int
cv_i(st_data_t k, st_data_t v, st_data_t a)
{
ID key = (ID)k;
st_table *tbl = (st_table *)a;
if (rb_is_class_id(key)) {
st_update(tbl, (st_data_t)key, cv_i_update, 0);
}
return ST_CONTINUE;
}
static void*
mod_cvar_at(VALUE mod, void *data)
{
st_table *tbl = data;
if (!tbl) {
tbl = st_init_numtable();
}
if (RCLASS_IV_TBL(mod)) {
st_foreach_safe(RCLASS_IV_TBL(mod), cv_i, (st_data_t)tbl);
}
return tbl;
}
static void*
mod_cvar_of(VALUE mod, void *data)
{
VALUE tmp = mod;
if (FL_TEST(mod, FL_SINGLETON)) {
if (rb_namespace_p(rb_ivar_get(mod, id__attached__))) {
data = mod_cvar_at(tmp, data);
tmp = cvar_front_klass(tmp);
}
}
for (;;) {
data = mod_cvar_at(tmp, data);
tmp = RCLASS_SUPER(tmp);
if (!tmp) break;
}
return data;
}
static int
cv_list_i(st_data_t key, st_data_t value, VALUE ary)
{
ID sym = (ID)key;
rb_ary_push(ary, ID2SYM(sym));
return ST_CONTINUE;
}
static VALUE
cvar_list(void *data)
{
st_table *tbl = data;
VALUE ary;
if (!tbl) return rb_ary_new2(0);
ary = rb_ary_new2(tbl->num_entries);
st_foreach_safe(tbl, cv_list_i, ary);
st_free_table(tbl);
return ary;
}
/*
* call-seq:
* mod.class_variables(inherit=true) -> array
*
* Returns an array of the names of class variables in <i>mod</i>.
* This includes the names of class variables in any included
* modules, unless the <i>inherit</i> parameter is set to
* <code>false</code>.
*
* class One
* @@var1 = 1
* end
* class Two < One
* @@var2 = 2
* end
* One.class_variables #=> [:@@var1]
* Two.class_variables #=> [:@@var2, :@@var1]
* Two.class_variables(false) #=> [:@@var2]
*/
VALUE
rb_mod_class_variables(int argc, const VALUE *argv, VALUE mod)
{
2019-11-25 09:05:53 +03:00
bool inherit = true;
st_table *tbl;
if (rb_check_arity(argc, 0, 1)) inherit = RTEST(argv[0]);
if (inherit) {
tbl = mod_cvar_of(mod, 0);
}
else {
tbl = mod_cvar_at(mod, 0);
}
return cvar_list(tbl);
}
/*
* call-seq:
* remove_class_variable(sym) -> obj
*
* Removes the named class variable from the receiver, returning that
* variable's value.
*
* class Example
* @@var = 99
* puts remove_class_variable(:@@var)
* p(defined? @@var)
* end
*
* <em>produces:</em>
*
* 99
* nil
*/
VALUE
rb_mod_remove_cvar(VALUE mod, VALUE name)
{
const ID id = id_for_var_message(mod, name, class, "wrong class variable name %1$s");
st_data_t val, n = id;
if (!id) {
goto not_defined;
}
rb_check_frozen(mod);
if (RCLASS_IV_TBL(mod) && st_delete(RCLASS_IV_TBL(mod), &n, &val)) {
return (VALUE)val;
}
if (rb_cvar_defined(mod, id)) {
rb_name_err_raise("cannot remove %1$s for %2$s", mod, ID2SYM(id));
}
not_defined:
rb_name_err_raise("class variable %1$s not defined for %2$s",
mod, name);
UNREACHABLE_RETURN(Qundef);
}
VALUE
rb_iv_get(VALUE obj, const char *name)
{
ID id = rb_check_id_cstr(name, strlen(name), rb_usascii_encoding());
if (!id) {
2019-07-01 08:00:37 +03:00
return Qnil;
}
return rb_ivar_get(obj, id);
}
VALUE
rb_iv_set(VALUE obj, const char *name, VALUE val)
{
ID id = rb_intern(name);
return rb_ivar_set(obj, id, val);
}
/* tbl = xx(obj); tbl[key] = value; */
int
rb_class_ivar_set(VALUE obj, ID key, VALUE value)
{
if (!RCLASS_IV_TBL(obj)) {
RCLASS_IV_TBL(obj) = st_init_numtable();
}
st_table *tbl = RCLASS_IV_TBL(obj);
int result = lock_st_insert(tbl, (st_data_t)key, (st_data_t)value);
RB_OBJ_WRITTEN(obj, Qundef, value);
return result;
}
static int
tbl_copy_i(st_data_t key, st_data_t value, st_data_t data)
{
RB_OBJ_WRITTEN((VALUE)data, Qundef, (VALUE)value);
return ST_CONTINUE;
}
void
rb_iv_tbl_copy(VALUE dst, VALUE src)
{
st_table *orig_tbl = RCLASS_IV_TBL(src);
st_table *new_tbl = st_copy(orig_tbl);
st_foreach(new_tbl, tbl_copy_i, (st_data_t)dst);
RCLASS_IV_TBL(dst) = new_tbl;
}
mjit_compile.c: merge initial JIT compiler which has been developed by Takashi Kokubun <takashikkbn@gmail> as YARV-MJIT. Many of its bugs are fixed by wanabe <s.wanabe@gmail.com>. This JIT compiler is designed to be a safe migration path to introduce JIT compiler to MRI. So this commit does not include any bytecode changes or dynamic instruction modifications, which are done in original MJIT. This commit even strips off some aggressive optimizations from YARV-MJIT, and thus it's slower than YARV-MJIT too. But it's still fairly faster than Ruby 2.5 in some benchmarks (attached below). Note that this JIT compiler passes `make test`, `make test-all`, `make test-spec` without JIT, and even with JIT. Not only it's perfectly safe with JIT disabled because it does not replace VM instructions unlike MJIT, but also with JIT enabled it stably runs Ruby applications including Rails applications. I'm expecting this version as just "initial" JIT compiler. I have many optimization ideas which are skipped for initial merging, and you may easily replace this JIT compiler with a faster one by just replacing mjit_compile.c. `mjit_compile` interface is designed for the purpose. common.mk: update dependencies for mjit_compile.c. internal.h: declare `rb_vm_insn_addr2insn` for MJIT. vm.c: exclude some definitions if `-DMJIT_HEADER` is provided to compiler. This avoids to include some functions which take a long time to compile, e.g. vm_exec_core. Some of the purpose is achieved in transform_mjit_header.rb (see `IGNORED_FUNCTIONS`) but others are manually resolved for now. Load mjit_helper.h for MJIT header. mjit_helper.h: New. This is a file used only by JIT-ed code. I'll refactor `mjit_call_cfunc` later. vm_eval.c: add some #ifdef switches to skip compiling some functions like Init_vm_eval. win32/mkexports.rb: export thread/ec functions, which are used by MJIT. include/ruby/defines.h: add MJIT_FUNC_EXPORTED macro alis to clarify that a function is exported only for MJIT. array.c: export a function used by MJIT. bignum.c: ditto. class.c: ditto. compile.c: ditto. error.c: ditto. gc.c: ditto. hash.c: ditto. iseq.c: ditto. numeric.c: ditto. object.c: ditto. proc.c: ditto. re.c: ditto. st.c: ditto. string.c: ditto. thread.c: ditto. variable.c: ditto. vm_backtrace.c: ditto. vm_insnhelper.c: ditto. vm_method.c: ditto. I would like to improve maintainability of function exports, but I believe this way is acceptable as initial merging if we clarify the new exports are for MJIT (so that we can use them as TODO list to fix) and add unit tests to detect unresolved symbols. I'll add unit tests of JIT compilations in succeeding commits. Author: Takashi Kokubun <takashikkbn@gmail.com> Contributor: wanabe <s.wanabe@gmail.com> Part of [Feature #14235] --- * Known issues * Code generated by gcc is faster than clang. The benchmark may be worse in macOS. Following benchmark result is provided by gcc w/ Linux. * Performance is decreased when Google Chrome is running * JIT can work on MinGW, but it doesn't improve performance at least in short running benchmark. * Currently it doesn't perform well with Rails. We'll try to fix this before release. --- * Benchmark reslts Benchmarked with: Intel 4.0GHz i7-4790K with 16GB memory under x86-64 Ubuntu 8 Cores - 2.0.0-p0: Ruby 2.0.0-p0 - r62186: Ruby trunk (early 2.6.0), before MJIT changes - JIT off: On this commit, but without `--jit` option - JIT on: On this commit, and with `--jit` option ** Optcarrot fps Benchmark: https://github.com/mame/optcarrot | |2.0.0-p0 |r62186 |JIT off |JIT on | |:--------|:--------|:--------|:--------|:--------| |fps |37.32 |51.46 |51.31 |58.88 | |vs 2.0.0 |1.00x |1.38x |1.37x |1.58x | ** MJIT benchmarks Benchmark: https://github.com/benchmark-driver/mjit-benchmarks (Original: https://github.com/vnmakarov/ruby/tree/rtl_mjit_branch/MJIT-benchmarks) | |2.0.0-p0 |r62186 |JIT off |JIT on | |:----------|:--------|:--------|:--------|:--------| |aread |1.00 |1.09 |1.07 |2.19 | |aref |1.00 |1.13 |1.11 |2.22 | |aset |1.00 |1.50 |1.45 |2.64 | |awrite |1.00 |1.17 |1.13 |2.20 | |call |1.00 |1.29 |1.26 |2.02 | |const2 |1.00 |1.10 |1.10 |2.19 | |const |1.00 |1.11 |1.10 |2.19 | |fannk |1.00 |1.04 |1.02 |1.00 | |fib |1.00 |1.32 |1.31 |1.84 | |ivread |1.00 |1.13 |1.12 |2.43 | |ivwrite |1.00 |1.23 |1.21 |2.40 | |mandelbrot |1.00 |1.13 |1.16 |1.28 | |meteor |1.00 |2.97 |2.92 |3.17 | |nbody |1.00 |1.17 |1.15 |1.49 | |nest-ntimes|1.00 |1.22 |1.20 |1.39 | |nest-while |1.00 |1.10 |1.10 |1.37 | |norm |1.00 |1.18 |1.16 |1.24 | |nsvb |1.00 |1.16 |1.16 |1.17 | |red-black |1.00 |1.02 |0.99 |1.12 | |sieve |1.00 |1.30 |1.28 |1.62 | |trees |1.00 |1.14 |1.13 |1.19 | |while |1.00 |1.12 |1.11 |2.41 | ** Discourse's script/bench.rb Benchmark: https://github.com/discourse/discourse/blob/v1.8.7/script/bench.rb NOTE: Rails performance was somehow a little degraded with JIT for now. We should fix this. (At least I know opt_aref is performing badly in JIT and I have an idea to fix it. Please wait for the fix.) *** JIT off Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 17 75: 18 90: 22 99: 29 home_admin: 50: 21 75: 21 90: 27 99: 40 topic_admin: 50: 17 75: 18 90: 22 99: 32 categories: 50: 35 75: 41 90: 43 99: 77 home: 50: 39 75: 46 90: 49 99: 95 topic: 50: 46 75: 52 90: 56 99: 101 *** JIT on Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 19 75: 21 90: 25 99: 33 home_admin: 50: 24 75: 26 90: 30 99: 35 topic_admin: 50: 19 75: 20 90: 25 99: 30 categories: 50: 40 75: 44 90: 48 99: 76 home: 50: 42 75: 48 90: 51 99: 89 topic: 50: 49 75: 55 90: 58 99: 99 git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62197 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-02-04 14:22:28 +03:00
MJIT_FUNC_EXPORTED rb_const_entry_t *
rb_const_lookup(VALUE klass, ID id)
{
struct rb_id_table *tbl = RCLASS_CONST_TBL(klass);
if (tbl) {
VALUE val;
bool r;
RB_VM_LOCK_ENTER();
{
r = rb_id_table_lookup(tbl, id, &val);
}
RB_VM_LOCK_LEAVE();
if (r) return (rb_const_entry_t *)val;
}
return NULL;
}