Segfault in tk/variable.rb

ruby-talk
1

Anybody else seen this? It's _extremely_ rare. I've seen it exactly twice,
both times running the same program - a test/unit test suite using
Tk::TestRunner. In the meantime I've run the program successfully dozens of
times. This last time it happened when I ran the program immediately after
issuing an `export -n ...' command.

tim: ~/vslick/rvg/test> export -n debug_prim=
tim: ~/vslick/rvg/test> ruby W3C*
/home/software/ruby-1.8.2//lib/ruby/1.8/tk/variable.rb:321: [BUG]
Segmentation fault
ruby 1.8.2 (2004-12-25) [i686-linux]

Aborted

2

Hi,

···

From: Tim Hunter <cyclists@nc.rr.com>
Subject: segfault in tk/variable.rb
Date: Thu, 13 Jan 2005 09:31:16 +0900
Message-ID: <ghjFd.16870$fE4.2512583@twister.southeast.rr.com>

tim: ~/vslick/rvg/test> export -n debug_prim=
tim: ~/vslick/rvg/test> ruby W3C*
/home/software/ruby-1.8.2//lib/ruby/1.8/tk/variable.rb:321: [BUG]
Segmentation fault
ruby 1.8.2 (2004-12-25) [i686-linux]

Could you tell me the value of TclTkLib::COMPILE_INFO ?
--
Hidetoshi NAGAI (nagai@ai.kyutech.ac.jp)

3

Hidetoshi NAGAI wrote:

Hi,

From: Tim Hunter <cyclists@nc.rr.com>
Subject: segfault in tk/variable.rb
Date: Thu, 13 Jan 2005 09:31:16 +0900
Message-ID: <ghjFd.16870$fE4.2512583@twister.southeast.rr.com>

tim: ~/vslick/rvg/test> export -n debug_prim=
tim: ~/vslick/rvg/test> ruby W3C*
/home/software/ruby-1.8.2//lib/ruby/1.8/tk/variable.rb:321: [BUG]
Segmentation fault
ruby 1.8.2 (2004-12-25) [i686-linux]

Could you tell me the value of TclTkLib::COMPILE_INFO ?

"tcltklib 2004-12-23 :: Ruby1.8.2 (2004-12-25) without pthread :: Tcl8.4.
(without stub)/Tk8.4.5(without stub) with tcl_threads"

4

Hi,

···

From: Tim Hunter <cyclists@nc.rr.com>
Subject: Re: segfault in tk/variable.rb
Date: Thu, 13 Jan 2005 11:46:21 +0900
Message-ID: <GhlFd.10248$hQ6.2371593@twister.southeast.rr.com>

> Could you tell me the value of TclTkLib::COMPILE_INFO ?

"tcltklib 2004-12-23 :: Ruby1.8.2 (2004-12-25) without pthread :: Tcl8.4.
(without stub)/Tk8.4.5(without stub) with tcl_threads"

Possibly, the trouble depends on inconsistency of pthread support.
Didn't you see the following messgase when compiling?

*****************************************************************************
**
** PTHREAD SUPPORT MODE ERRROR:
**
** Ruby is not compiled with --enable-pthread, but your Tcl/Tk
** libararies seems to be compiled with "pthread support". This
** combination possibly cause "Hang-up" or "Segmentation Fault"
** frequently when Ruby/Tk is working. We NEVER recommend you to
** create the library under such combination of pthread support.
**
** Please recompile Ruby with "--enable-pthread" configure option
** or recompile Tcl/Tk with "--disable-threads" configure option.
**
*****************************************************************************

--
Hidetoshi NAGAI (nagai@ai.kyutech.ac.jp)

5

Hidetoshi NAGAI wrote:

···

Hi,

From: Tim Hunter <cyclists@nc.rr.com>
Subject: Re: segfault in tk/variable.rb
Date: Thu, 13 Jan 2005 11:46:21 +0900
Message-ID: <GhlFd.10248$hQ6.2371593@twister.southeast.rr.com>

> Could you tell me the value of TclTkLib::COMPILE_INFO ?

"tcltklib 2004-12-23 :: Ruby1.8.2 (2004-12-25) without pthread :: Tcl8.4.
(without stub)/Tk8.4.5(without stub) with tcl_threads"

Possibly, the trouble depends on inconsistency of pthread support.
Didn't you see the following messgase when compiling?

*****************************************************************************

**
** PTHREAD SUPPORT MODE ERRROR:
**
** Ruby is not compiled with --enable-pthread, but your Tcl/Tk
** libararies seems to be compiled with "pthread support". This
** combination possibly cause "Hang-up" or "Segmentation Fault"
** frequently when Ruby/Tk is working. We NEVER recommend you to
** create the library under such combination of pthread support.
**
** Please recompile Ruby with "--enable-pthread" configure option
** or recompile Tcl/Tk with "--disable-threads" configure option.
**

*****************************************************************************

Thank you for helping me.

Hmmm...Why is the default for Ruby different from Tcl/Tk's default when I'm
building them at the same time? I did not use --enable-pthread when
building Ruby, so how did Tcl/Tk get --enable-threads?

To be sure, I did a `make clean' and re-ran ./configure without the
--enable-pthread option. When I ran `make' I did not get this message.

Then I ran ./configure with --enable-pthread. When I ran `make' I _did_ get
the message.

Which is right?

6

Message-ID: <TiDFd.16990$fE4.2766628@twister.southeast.rr.com>

Hmmm...Why is the default for Ruby different from Tcl/Tk's default when I'm
building them at the same time? I did not use --enable-pthread when
building Ruby, so how did Tcl/Tk get --enable-threads?

Oh, I'm verry sorry. That is a bug of tcltklib/extconf.rb.
When disable-pthread, the logical value of WITH_TCL_ENABLE_THREAD macro
is inverted.
Unless "ruby -r tk -e 'p Tk::PLATFORM["threaded"]'" shows "1",
your tcl/tk which linked your ruby is NOT enable-threads.

To be sure, I did a `make clean' and re-ran ./configure without the
--enable-pthread option. When I ran `make' I did not get this message.

Then I ran ./configure with --enable-pthread. When I ran `make' I _did_ get
the message.

Probably, that is a warning message of inconsistency, isn't that?
When ruby=>pthread-enable and tcl/tk=>thread-disable,
output the warning message but succeed to compile.

Well, if there are no consistency problems,
I have to search the reason of your trouble.
tk/variable.rb.321 is

···

From: Tim Hunter <cyclists@nc.rr.com>
Subject: Re: segfault in tk/variable.rb
Date: Fri, 14 Jan 2005 08:16:30 +0900
---------------------------------------------------------------------------
      _fromUTF8(INTERP._set_global_var(@id, _get_eval_string(val, true)))
---------------------------------------------------------------------------
I want to know which of _fromUTF8, _set_global_var and
_get_eval_string raises SEGV.
If you don't mind, could you check and report it?
--
Hidetoshi NAGAI (nagai@ai.kyutech.ac.jp)

7

Hidetoshi NAGAI wrote:

···

Well, if there are no consistency problems,
I have to search the reason of your trouble.
tk/variable.rb.321 is

---------------------------------------------------------------------------

      _fromUTF8(INTERP._set_global_var(@id, _get_eval_string(val, true)))

---------------------------------------------------------------------------

I want to know which of _fromUTF8, _set_global_var and
_get_eval_string raises SEGV.
If you don't mind, could you check and report it?

I don't mind at all. Since the segv occurs only very rarely, it may be some
time before I see it again. If it happens again I'll follow up.

Thanks again for advising me.

8

Hi,

···

From: Tim Hunter <cyclists@nc.rr.com>
Subject: Re: segfault in tk/variable.rb
Date: Sat, 15 Jan 2005 08:56:12 +0900
Message-ID: <9_YFd.18554$fE4.3108651@twister.southeast.rr.com>

Hidetoshi NAGAI wrote:
> Well, if there are no consistency problems,
> I have to search the reason of your trouble.
> tk/variable.rb.321 is
>
---------------------------------------------------------------------------
> _fromUTF8(INTERP._set_global_var(@id, _get_eval_string(val, true)))
>
---------------------------------------------------------------------------
> I want to know which of _fromUTF8, _set_global_var and
> _get_eval_string raises SEGV.
> If you don't mind, could you check and report it?

I don't mind at all. Since the segv occurs only very rarely, it may be some
time before I see it again. If it happens again I'll follow up.

Possibly fixed.
I thought the problem depends on TclTkIp#_set_global_var.
When the method calls the Tk interpreter during the Tk interpreter
is being deleted, sometimes the current namespace of Tk interpreter
points NULL.
I added the current namespace check at checking deleted interpreter.
I've committed the fix. Please see the latest CVS.
--
Hidetoshi NAGAI (nagai@ai.kyutech.ac.jp)