[ruby-core:71164] [Ruby trunk - Feature #11614] [Open] [RFC] use id_table for constant tables

[ruby-core:71164] [Ruby trunk - Feature #11614] [Open] [RFC] use id_table for constant tables

[normalperson]

Issue #11614 has been reported by Eric Wong.

----------------------------------------
Feature #11614: [RFC] use id_table for constant tables
https://bugs.ruby-lang.org/issues/11614

* Author: Eric Wong
* Status: Open
* Priority: Normal
* Assignee: 
----------------------------------------
ref: [ruby-core:71142] [ruby-core:71151]

I'm not sure if using rb_class_name in object.c is enough to prevent
breaking all existing code.  Until this is sufficiently reviewed,
I have no plans to commit this patch at the moment.

valgrind 3.9.0 on x86-64 reports a minor reduction in memory usage
when loading only RubyGems and RDoc by running: ruby -rrdoc -eexit

before: HEAP SUMMARY:
    in use at exit: 2,913,448 bytes in 27,394 blocks
  total heap usage: 48,362 allocs, 20,968 frees, 9,034,621 bytes alloc

after: HEAP SUMMARY:
    in use at exit: 2,880,056 bytes in 26,712 blocks
  total heap usage: 47,791 allocs, 21,079 frees, 9,046,507 bytes alloc

* class.c (struct clone_const_arg): adjust for id_table
  (clone_const): ditto
  (clone_const_i): ditto
  (rb_mod_init_copy): ditto
  (rb_singleton_class_clone_and_attach): ditto
  (rb_include_class_new): ditto
  (include_modules_at): ditto
* constant.h (rb_free_const_table): ditto
* gc.c (free_const_entry_i): ditto
  (rb_free_const_table): ditto
  (obj_memsize_of): ditto
  (mark_const_entry_i): ditto
  (mark_const_tbl): ditto
* internal.h (struct rb_classext_struct): ditto
* object.c (rb_mod_const_set): resolve class name on assignment
* variable.c (const_update): replace with const_tbl_update
  (const_tbl_update): new function
  (fc_i): adjust for id_table
  (find_class_path): ditto
  (autoload_const_set): st_update => const_tbl_update
  (rb_const_remove): adjust for id_table
  (sv_i): ditto
  (rb_local_constants_i): ditto
  (rb_local_constants): ditto
  (rb_mod_const_at): ditto
  (rb_mod_const_set): ditto
  (rb_const_lookup): ditto


---Files--------------------------------
0001-use-id_table-for-constant-tables.patch (18.3 KB)


-- 
https://bugs.ruby-lang.org/

[ruby-core:72112] Re: [Ruby trunk - Feature #11614] [Open] [RFC] use id_table for constant tables

[Eric Wong]

normalperson@yhbt.net wrote:
> Feature #11614: [RFC] use id_table for constant tables
> https://bugs.ruby-lang.org/issues/11614

Any comment?  I'm pretty sure this was alright,
but maybe it's 2.4 material.  Thanks.

Also related: [ruby-core:71455]
<http://blade.nagaokaut.ac.jp/cgi-bin/scat.rb/ruby/ruby-core/71455>
<http://80x24.org/spew/20151111201126.22805-1-e%4080x24.org/raw>
for iv_index_tbl, but that's more likely for 2.4.
ID_TABLE_USE_ID_SERIAL limits the flexibility of id_table greatly :/

[ruby-core:72484] Re: [Ruby trunk - Feature #11614] [Open] [RFC] use id_table for constant tables

[Eric Wong]

Eric Wong <normalperson@yhbt.net> wrote:
> normalperson@yhbt.net wrote:
> > Feature #11614: [RFC] use id_table for constant tables
> > https://bugs.ruby-lang.org/issues/11614
> 
> Any comment?  I'm pretty sure this was alright,
> but maybe it's 2.4 material.  Thanks.

Updated patch on top of latest trunk r53315

http://80x24.org/spew/20151226073256.24943-1-e%4080x24.org/raw

I think I'll commit in a few days since we have a year until 2.4