From: "jonforums (Jon Forums)" <redmine@...>
Date: 2013-02-21T12:31:39+09:00
Subject: [ruby-core:52608] [ruby-trunk - Bug #5199] ext/tk: RB_GC_GUARD seems to be needed in several places


Issue #5199 has been updated by jonforums (Jon Forums).


nobu (Nobuyoshi Nakada) wrote:
> Quick patch.
> 
> 2.0 has GC safe functions, rb_sprintf() with PRIsVALUE, rb_intern_str(), and so on.

Does a list of these new 2.0 GC-safe functions exist anywhere?

If the 2.0 GC-safe functions aren't currently enumerated in a single, accessible place, should they be listed in a new section to README.EXT, and a short note added to mame-san's draft 2.0.0-p0 ANN? Perhaps under "Built-in libraries" as "New GC-safe API for more robust extensions."
----------------------------------------
Bug #5199: ext/tk: RB_GC_GUARD seems to be needed in several places
https://bugs.ruby-lang.org/issues/5199#change-36695

Author: normalperson (Eric Wong)
Status: Assigned
Priority: Normal
Assignee: nagai (Hidetoshi Nagai)
Category: ext
Target version: 2.1.0
ruby -v: ruby 1.9.4dev (2011-08-11 trunk 32931) [x86_64-linux]


I noticed the following in ext/tk/tkutil/tkutil.c:

    rb_warning("fail to convert '%s' to string for Tk",
                   RSTRING_PTR(rb_funcall(obj, rb_intern("inspect"), 0,
                   0)));

    id = rb_intern(RSTRING_PTR(rb_str_cat2(rb_str_new2("@"), str)));

I don't use or know the Tk ext at all, but the above examples do not
appear safe from a GC perspective.  Compilers can optimize the original
VALUE away entirely so GC can collect.

RSTRING_PTR (being a macro,) may also evaluate its arguments multiple
times.



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