On Fri, May 06, 2016 at 11:23:38AM +0200, Bill Allombert wrote:
> On Thu, May 05, 2016 at 01:24:32PM -0700, John Jones wrote:
> > Package: pari
> > Version: latest from git
> >
> > This bug is intermittent.  I have had recent problems with other functions,
> > so it may be a memory management issue, or something with variables?
> >
> > Here is a transcript of a session.  After defining my function (sorry it is
> > long), I run the function several times with the same inputs.  The function
> > is computing quadratic extensions unramified outside a given set.  Anyway,
> > the first few times I run it in the session below, it fails on one
> > congruence subgroup, and then it works.  When it fails, rnfkummer has
> > returned 0.
> >
> > Hobbes ~/gp_progs: gp -f
> >
> >           GP/PARI CALCULATOR Version 2.8.0 (development 18363-3cbc853)
>
> Hello John, 3cbc853 is from last year, it is not the latest GIT version.
> I can reproduce this ith 3cbc853 but not with a42509159.
>
> I will do a bisection.

It seems it was fixed in

commit c3956403b8c4dabe56683a49dca2f55fabba27c9
Author: Karim Belabas <Karim.Belabas@math.u-bordeaux1.fr>
Date:   Thu Jan 14 08:37:23 2016 +0100

    partially revert 4915686ad [#1776]

    The call in ArtinNumber needs more properties of
set_sign_mod_divisor
    than are documented. To be improved.

Cheers,
Bill.