Fwd: help with debugging

[Adrian Buehlmann]

On 18.02.2010 10:16, varol kaptan wrote:
> Sorry, the following reply was supposed to go to the list
> 
> ---------- Forwarded message ----------
> From: varol kaptan <varol.kaptan at gmail.com>
> Date: Wed, Feb 17, 2010 at 6:10 PM
> Subject: Re: help with debugging
> To: Patrick Mézard <pmezard at gmail.com>
> 
> 
> On Wed, Feb 17, 2010 at 2:52 PM, Patrick Mézard <pmezard at gmail.com> wrote:
>> Le 17/02/10 00:51, varol kaptan a écrit :
>>> Hi,
>>>
>>> I have a weird problem:
>>>
>>> Mercurial 1.4.3 / TortoiseHG 0.9.3, no extensions, winxp.
>>>
>>> I am trying to commit a change in one file (~ 10Mb xls on winxp) and
>>> after about 5-10 seconds the command prompt comes back without any
>>> error message. A subsequent "hg status" reveals that the commit did
>>> not happen. There is an unfinished transaction and a wlock file in
>>> .hg/ (If I run the commit command again, mercurial advises me to run
>>> hg recover first.)
>>>
>>> I've tried running hg with -v --debug --traceback without any success.
>>> Something obviously goes wrong at some point but nothing is displayed
>>> on the screen. The same commit succeeds on a linux box with Mercurial
>>> 1.2.1 if I copy the directory and remove the lock file(s). Version
>>> 1.2.1 on the xp box does not work either - same as 1.4.3.
>>>
>>> Any suggestions on how to diagnose such a problem?
>>
>> Do you see anything like a memory spike during the commit?
>>
>>
>> Also, you can find source/development based installers here:
>>
>>    http://bitbucket.org/tortoisehg/thg-winbuild/downloads/
>>
>> For instance,
>>
>>    http://bitbucket.org/tortoisehg/thg-winbuild/downloads/mercurial-1.4.3.win32-py2.6.exe
>>
>> requires you to have a local python 2.6 install. The benefit is you will be running python code directly and have the ability to hack the .py files. Plus the reporting functions may work better.
>>
>> --
>> Patrick Mézard
>>
> 
> Thank you very much.
> 
> it turns out that this is issue1940 - stack overflow in bdiff.c -
> replacing with bdiff.py (the pure python implementation) fixes the
> problem.
> 
> I am surprised that to see that this patch has not made it yet (see
> last post at http://markmail.org/thread/fjz2arydwo7skjki )
> 

Interesting.

Would be interesting to know if using a build of hg using the Microsoft C
compiler still has this problem (Without applying the mentioned patch.
Does MS C optimize better?).

I guess mercurial-1.4.3.win32-py2.6.exe was built using MinGW C compiler.

mercurial-stable-1.4.3+168-dc9a803c1800.msi from
http://bitbucket.org/tortoisehg/thg-winbuild/downloads/ should
have been built with the MS C compiler.

Warning: if you install tortoisehg*-{x86|x64}.msi from there, you
need to uninstall 0.9.3 first. BEFORE DOING SO BACKUP YOUR USER MERCURIAL.INI
(it might get nuked by the 0.9.3 UNinstaller).

See also
http://www.mail-archive.com/tortoisehg-discuss@lists.sourceforge.net/msg01929.html