Discussion related to cado-nfs

[Emmanuel Thomé <emmanuel.thome@gmail.com>]

On Fri, Jul 19, 2013 at 3:33 PM, meng <qsmeng@126.com> wrote:
>>/home/greatnet/cado-nfs/build/master/sieve/las -I 11 -rpowlim 2047 \
>> -apowlim 2047 -poly /tmp/cado.QQVgf15501/c59.poly \
>> -fb /tmp/cado.QQVgf15501/c59.roots -q0 200000 -q1 202000 -mt 4 \
>> -out /tmp/cado.QQVgf15501/c59.rels.200000-202000.gz
[...]
> *** glibc detected *** /home/greatnet/meng/cado-nfs/build/master/sieve/las:
> free(): invalid next size (fast): 0x00002ad49c000fe0 ***


Hi.

I fear that redhat el5 with glibc-2.5 and gcc-4.1.2 might be a
severely outdated setup. This does not forbid trying to understand
what's happening, though.

There are a couple of things that can happen.
 - bug in cado-nfs. As far as I can tell, you did not tell which
version of cado-nfs you were using. If you are still using
cado-nfs-1.1, please update to the git version, so that we don't waste
time diagnosing a problem which has disappeared.

 - system and/or compiler bug. Miscompilations do happen. I don't
recall whether gcc-4.1.2 was on the blacklist or not, but I do recall
that some oldish libcs had a problem with threads which impacted
cado-nfs. You can try to compile with a more recent gcc, for instance.
You may also try to run the above command without the "-mt 4"
parameter.

 - discrepancy between the machine on which you compiled cado-nfs and
where you ran it, although this case does not seem likely, given the
error message.

Another thing you can do is to compile with debug symbols, and
valgrind the thing (assuming valgrind is installed):

rm -rf build/master
CFLAGS="-g -W -Wall" CXXFLAGS="-g -W -Wall" make -j8 las
valgrind /home/greatnet/cado-nfs/build/master/sieve/las -I 11 -rpowlim
2047 [...]

(again, with "-mt 4" removed).

We might be interested in the output.

Best,

E.