Re: valgrinding with python stuff active

On 08/09/2012 17:52, Norbert Thiebaud wrote:
> On Fri, Sep 7, 2012 at 5:49 PM, julien2412<serval2412@yahoo.fr>  wrote:
> > on this page http://wiki.documentfoundation.org/Development/How_to_debug
> > but what about if custom memory allocator is bugged? (BTW why do we use
> > this? Historical reason?) Wouldn't Valgrind be helpful to detect it?
> >
> > I'm a bit lost here :-(
> Julien,
>
> A custom allocator will necessarily confuse valgrind...  the only way
> to 'unconfuse' it would be to teach valgrind about it and have
> valgrind emulate the allocation mechanism of the custom allocator to
> keep track of what is 'logically' allocated/freed... doing that would
> be quite complex and the odd of doing a bug in the emulation layer is
> much higher than the odd that there is a bug in the custom allocator
> to start with...
>
> So, in general, when using valgrind you _do_ want to disable any
> 'custom allocator'. and yes that won't help you find a bug in the
> allocator... but if you have a bug, think horse not zebra... 99.999%
> of the time your bug will have nothing to do with the allocator
> itself.
>
> Norbert
Thank you Norbert for your answer, it's clearer for me now. (moreover I 
didn't know this expression, "think horse not zebra" :-))
So in 
http://wiki.documentfoundation.org/BugReport#How_to_get_valgrind_log_.28on_Linux.29 
page, should I add this:

export G_SLICE=always-malloc

before this:

valgrind --tool=memcheck --num-callers=50 --trace-children=yes ./soffice.bin 2>&1 | tee 
/tmp/valgrind.log

?

Julien