pygoocanvas causes immediate segfault on import on x86-64

[Adam Williamson]

More info will be needed to fix this but I don't really know how to debug stuff in Python so I'm going to need instructions from you on how to provide the extra info.

Building Conduit, I noticed it failed on our (Mandriva's) x86-64 buildhost but succeeded on i586. It was failing on the pygoocanvas test, which just tries to import pygoocanvas. Further inspection reveals:

x86-64

[iurt cooker] adamwklodia~$ python
Python 2.5.2 (r252:60911, Aug 25 2008, 10:52:07)
[GCC 4.3.1 20080626 (prerelease)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import goocanvas
Segmentation fault

i586

[adamw@lenovo SPECS]$ python
Python 2.5.2 (r252:60911, Aug 25 2008, 09:58:25)
[GCC 4.3.1 20080626 (prerelease)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import goocanvas
>>>

That's all I have so far. The builds are identical, the only difference is the arch (i586 vs. x86-64). How do I go about getting the necessary info to let you fix this? Thanks!

[Gian Mario Tagliaretti (gianmt)]

Hi Adam,

thanks for your bug report. can you tell me what version of pygobject are you using?

We recently had a similar problem in pygtk caused by a bug in pygobject (up to 2.15.3), it's now fixed in trunk.

cheers

[Adam Williamson (awilliamson)]

On Wed, 2008-09-03 at 09:20 +0000, Gian Mario Tagliaretti wrote:
> Hi Adam,
>
> thanks for your bug report. can you tell me what version of pygobject
> are you using?
>
> We recently had a similar problem in pygtk caused by a bug in pygobject
> (up to 2.15.3), it's now fixed in trunk.

Ah, I see. We are indeed on python-gobject 2.15.3.

Is this fixed in pygobject trunk or pygoocanvas trunk? (And where do I
find pygoocanvas trunk anyhow?)
--
adamw

[Gian Mario Tagliaretti (gianmt)]

On Wed, Sep 3, 2008 at 5:01 PM, Adam Williamson
<email address hidden> wrote:
> On Wed, 2008-09-03 at 09:20 +0000, Gian Mario Tagliaretti wrote:

>> We recently had a similar problem in pygtk caused by a bug in pygobject
>> (up to 2.15.3), it's now fixed in trunk.

> Ah, I see. We are indeed on python-gobject 2.15.3.
> Is this fixed in pygobject trunk or pygoocanvas trunk?

Has been fixed in pygobject trunk, and a 2.15.4 release has been issued.

> (And where do I
> find pygoocanvas trunk anyhow?)

http://svn.gnome.org/viewvc/pygoocanvas/

[Gian Mario Tagliaretti (gianmt)]

I close as invalid since the problem it's in pygobject, please feel free to reopen if upgrading to 2.15.4 doesn't fix the problem.

[Adam Williamson (awilliamson)]

Right, while I was asleep someone already updated our package to 2.15.4, and now it works. Thanks!

[Tolgam (tolgam)]

Hum... I'm sorry but I had the same problem and getting pygobject to 2.15.4 did not solved it :-/

[Tolgam (tolgam)]

Just tried again using both pygobject trunk and pygoocanvas trunk; same result :(

[Gustavo Carneiro (gjc)]

I have x86_64 and cannot reproduce your problem. Even valgrind doesn't catch anything:
gjc@dark-tower:Desktop$ valgrind python -c "import goocanvas"
==7691== Memcheck, a memory error detector.
[...]
==7691== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 155 from 1)

Could you run with valgrind, like I did above?

[Tolgam (tolgam)]

Yup ! I ran valgrind python -c "import goocanvas" > valgrind.log 2>&1
The result is attached.

[Gustavo Carneiro (gjc)]

Error here:

==27756== Jump to the invalid address stated on the next line
==27756== at 0x0: ???
==27756== by 0x6292088: PyEval_GetGlobals (in /usr/lib/python2.6/site-packages/goocanvasmodule.so)
==27756== by 0x62AF922: PyImport_Import (in /usr/lib/python2.6/site-packages/goocanvasmodule.so)
==27756== by 0x62AFB74: PyImport_ImportModule (in /usr/lib/python2.6/site-packages/goocanvasmodule.so)
==27756== by 0x62313DD: PyCObject_Import (in /usr/lib/python2.6/site-packages/goocanvasmodule.so)
==27756== by 0x620ADD5: initgoocanvas (goocanvasmodule.c:62)

goocanvasmodule.c line 62 contains:

    Pycairo_IMPORT;

Does 'import cairo' work?

[Tolgam (tolgam)]

Yes, it works. Log from valgrind also attached (in case it would be usefull anyway)

[Gustavo Carneiro (gjc)]

Alas, in that case I cannot offer any more hypotheses. Pycairo_IMPORT; expands to:

   Pycairo_CAPI = (Pycairo_CAPI_t*) PyCObject_Import("cairo", "CAPI");

If this crashes, it is deep inside Python. Clearly, pygoocanvas is not doing anything wrong.

I don't really know how to debug this further without forcing you to recompile from scratch python itself (with debugging symbols), pycairo, pygobject, pygtk, and finally pygoocanvas.

Alternatively, it's a long shot, but perhaps the output of the following command will offer some clue:

   strace python -c "import goocanvas"

[Gian Mario Tagliaretti (gianmt)]

Any news on this one?

[Douglas Soares de Andrade (dsandrade)]

Hi,

Im having this problem in an ArchLinux i686 system, here is the details:

- Archlinux i686
- Packages

goocanvas 0.13-1
pygoocanvas 0.13.1-1
pygtk 2.13.0-2
pygobject 2.16.0-1
cairo 1.8.6-1
pycairo 1.8.2-1
python 2.6.1

When i do python -c "import goocanvas" it gives me a segfault. I did not saw anything in valgrind and in strace, but if needed i can send the logs.

Im asking this because im the pygoocanvas maintainer in archlinux and it is not working, so i had to do a downgrade (0.12.0) to make it work.

Thanks a lot