python-sybase-misc

[python-sybase] Coredumps with Python 2.5 / Solaris 10

[Gregory B. <gn...@it...>]

I'm really struggling with making the 0.37 module work with Python 2.5.  I'm
getting some SIGSEGVs buried deep in the Python core, usually in the
realloc/free calls, in all sorts of internal Python objects (e.g.
PyList_Append(), PyDict_SetItem() etc) that really orta be very well tested
and debugged!

If I configure Python with either --with-pydebug or --without--pymalloc
options (i.e. Changing the way objects/mallocs work) then the resulting
Python version works fine with the Sybase module.  SunPro or GCC compiled,
both fail.

Python 2.4.4 works fine with no special config options.

All of which is telling me something is corrupting heap memory.  Some notes
in the 2.5 release notes indicate the memory mgmt for C extensions has been
altered but as far as I can see, the sybase module doesn't use tiehr the
PyMem_Malloc or PyObject_Malloc calls discussed there.

Has anyone else succeeded in using the Sybase module with Python 2.5?

Sparc / Solaris 10 
Sybase 12_5_3 libraries

Re: [python-sybase] Coredumps with Python 2.5 / Solaris 10

[Andrew M. <an...@ob...>]

>All of which is telling me something is corrupting heap memory.  Some notes
>in the 2.5 release notes indicate the memory mgmt for C extensions has been
>altered but as far as I can see, the sybase module doesn't use tiehr the
>PyMem_Malloc or PyObject_Malloc calls discussed there.

What makes these failures difficult to debug is that the crash happens
away from the source of the corruption, so anything that closes that
gap will help isolate the source.

In the past, I've used techniques such as a "canary" function that
performs a series of randomly sized allocations and then frees them, in
the hope that this will trip up on the corruption. I then strategically
litter the code with calls to this function. It might also help if the
canary function writes a "hard to swallow" pattern, such as 0xA5A5A5A5,
to the memory it allocates prior to freeing it.

-- 
Andrew McNamara, Senior Developer, Object Craft
http://www.object-craft.com.au/

Re: [python-sybase] Coredumps with Python 2.5 / Solaris 10

[Harri P. <har...@tr...>]
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[PATCH] Re: [python-sybase] Coredumps with Python 2.5 / Solaris 10

[Gregory B. <gn...@it...>]

On 20/10/06 12:46 PM, "Gregory Bond" <gn...@it...> wrote:

> I'm really struggling with making the 0.37 module work with Python 2.5.  I'm
> getting some SIGSEGVs buried deep in the Python core, usually in the
> realloc/free calls, in all sorts of internal Python objects (e.g.
> PyList_Append(), PyDict_SetItem() etc) that really orta be very well tested
> and debugged!
> 
> If I configure Python with either --with-pydebug or --without--pymalloc
> options (i.e. Changing the way objects/mallocs work) then the resulting
> Python version works fine with the Sybase module.  SunPro or GCC compiled,
> both fail.
> 
> Python 2.4.4 works fine with no special config options.
> 
> All of which is telling me something is corrupting heap memory.  Some notes
> in the 2.5 release notes indicate the memory mgmt for C extensions has been
> altered but as far as I can see, the sybase module doesn't use tiehr the
> PyMem_Malloc or PyObject_Malloc calls discussed there.

Found it. (I think..... Still a bit new to the Python extending caper.)

The varous *_dealloc() functions in sybasect were using PyMem_DEL() to
release memory.  Under 2.5, this needs to be PyObject_DEL() as the default
2.5 Python allocator keeps these in separate pools.

Patch attached!