Bug#780186: GMP doesn't manage memory correctly

[Julien Puydt]

Hi,

Le lundi 08 juin 2015 à 22:26:27 (-0500), Steve M. Robbins a écrit :
> On Tue, Mar 10, 2015 at 09:07:52AM +0100, Julien Puydt wrote:
> > Package: gmp
> > Version: 2:6.0.0+dfsg-6
> > 
> > Hi,
> > 
> > the sagemath software is using MPIR, but we're trying to package it for
> > debian using the default lib for multi-precision arithmetic : GMP.
> > 
> > Sagemath uses mp_set_memory_functions to override the default memory
> > management and make it better for their purpose (trigger their error
> > handling), but there are places where gmp still calls its own functions!
> 
> Please list the places.

Uh... I don't remember at all -- I'm still trying to package sagemath,
but since I had no answer I focused on other problems...

> > And we have a failing test in their suites where an abort is triggered,
> > because _mpz_realloc in mpz/realloc.c is called instead of the
> > sagemath-provided realloc.
> 
> "their" suites -- is that a GMP test suite or something else?

Ah, I had forgotten the "test" word ; indeed that was referring to
sagemath's own test suite.
 
> > GMP shouldn't call its own realloc if a better one has been provided.
> 
> I'll need more information to process this bug further.

I'm unable to provide it any more unfortunately :-(

Snark on #debian-python and #debian-science