Bug#764510: gcc-4.9: gnucash segfaults when built with GCC 4.9

[Matthias Klose <doko@debian.org>]

Am 09.10.2014 um 17:00 schrieb Frédéric Brière:
> [ I assumed the BTS would send a Cc to the gnucash maintainer due to the
> affects header, but that probably wasn't the case.  Sorry, Sébastien. ]
> 
> On Wed, Oct 08, 2014 at 08:25:05PM +0200, Matthias Klose wrote:
>> are there any suspicious compiler warnings? maybe fix these first.
> 
> -Werror -Wall passes with flying colors.
> 
>> please try to build gnucash with -O1 or -O0 and see if it works. If yes, build
>> half of the object files with -O2 (the default) and with -O0.  Then try to
>> reduce the files built with -O0.  Once you have identified one file, continue
>> using the optimize attribute to lower the optimization level for a function
>> until you have identified the misbuilt function.
> 
> Thanks!  That was exactly what I needed to track it down.
> 
> As inferred by the backtrace from my previous email, the culprit is
> indeed gsslrtma_populate_tree_store(); optimizing it with -O1 or -O2
> triggers the segfault.
> 
> I tried to dig a little deeper, but to no avail.  The collect_values
> argument to value_collect_string() on frame 2 is "bad", but inbetween
> that and gsslrtma_populate_tree_store() are three frames full of
> optimized-out variables, opaque varargs, and internal GTK/GNOME macros.
> That's beyond my league.
> 
> 
> Sébastien, I think I'll pass the baton to you now.  :) Let me know if
> there's anything I can do to help.

please attach the .i and .s files, for both the unmodified version, and the
version, where you added the attribute to this function, plus the command line
options how to build these files.