On Sat, Apr 05, 2014 at 11:13:17AM +0200, Bill Allombert wrote:
On Sat, Apr 05, 2014 at 09:46:36AM +0200, Xavier Roblot wrote:
Package: PARI
Version: development   16306-3443337

Hi,

The nfields part of make bench is failing for gp-dyn on my OS X 10.9.2 system with a bunch of SEGFAULT. On the other hand, gp-sta runs just fine. I attach the corresponding output file. The first error occurs with the function bnfcompress.

The first bad commit is

commit 8bacf11ac11df738533edb64f8a8f040468b581f
Author: Bill Allombert <Bill.Allombert@math.u-bordeaux1.fr>
Date:   Mon Mar 19 19:05:29 2012 +0100

   Replace pari_init_stack by paristack_alloc/paristack_resize.

I can reproduce this on mammouth.
However the bug seems to have disappeared in master.
The first good commit is 78b068c6bdff07169e125e0bca8db3e7f79c1a36

I can confirm that the bug has disappeared.

Also we really reommend to use gp-sta when using --mt=pthread.
Even on Linux, there is still issues with TLS support in dynamic libraries
that can cause gp-dyn to segfault when recovering from errors.

I agree with this but, by default, it is the dynamic gp that is installed in /bin, it appears. I do not know how to change this behaviour. For the moment, I use a link to the static version.