Re: [MiNT] Malloc Bug

[Paul Wratt <paul.wratt@gmail.com>]

On Mon, Jan 25, 2010 at 3:01 PM, Mark Duckworth
<mduckworth@atari-source.org> wrote:
> On 1/25/10 3:53 PM, Mark Duckworth wrote:
>>
>> Hey guys,
>>
>> More detail on the malloc bug.  I've been able to almost consistently
>> reproduce it on my gcc 4.4.x/cvs mintlib aranym system.  The package that
>> most consistently produces it is an rpm --rebuild of rcs.  It dies like
>> this:
>>
>> /usr/include/ctype.h:76:1: warning: this is the location of the previous
>> definition
>> gcc  -g -O ident.o rcsmap.o version.o  -o ident
>> gcc -c  -Dhas_conf_h  -I. -I. -g -O merge.c
>> make: ../mintlib/malloc.c:187: __free: Assertion `s == r' failed.
>> make: *** [install] Error 134
>> Bad exit status from /var/tmp/rpm-tmp.54754 (%install)
>> root@arawork:/h/srpms>
>>
>> Importantly, it has done this before, and now it is doing this every time
>> I try.  I'm assuming that if I reboot the problem will be gone.  If it's not
>> then we're really talking.  But on this system I have done NOTHING more than
>> start xaaes and do package builds so the problem lies within those
>> facilities.  Next time I will not start xaaes.
>>
>> Thanks,
>> Mark
>>
>>
>>
> Yeah it doesnt' do it after a fresh boot.  But what I will do is rebuild
> make with debugging symbols, build a  bunch of packages and revisit it.
>  Maybe I can examine the buffers and see what is in there.  Maybe something
> will stand out.
>
> Thanks,
> Mark
>
anything to try and get a consistant reproduction

So I gather so far, you need something that splinters the memory in
some way, to the point where an rpm --rebuild of rcs will consistantly
fail.

I suggest using zView on a large images folder, paired with FreeGEM 2
to quickly get the desired effect. You will need a large image folder
tho.

Hmm, ARAnyM with a "save state" might be useful here. If someone could
produce the problem in STEem (or any emu with state saving) this might
speed up the finding of a solution to this malloc issue..

Paul