On 27 July 2015 at 11:45, Anders Logg <anders.l...@gmail.com> wrote: > I mean it looks very specific and that the number of iterations is likely > to change if there is an upstream improvement or regression to the > preconditioner or if the logic of checking the convergence criterion is > slightly modified. > > The reference number of iterations is computed at the beginning of the test (it's not hard-wired), so it's robust with respect to all the above issues. The test failure is showing that (a) the old version of PETSc has a bug, or (b) that DOLFIN is not interfacing properly to an old version of PETSc. Neither case is worth fixing.
Garth > > mån 27 juli 2015 kl 12:42 skrev Garth N. Wells <gn...@cam.ac.uk>: > >> On 27 July 2015 at 11:37, Anders Logg <anders.l...@gmail.com> wrote: >> >>> Yes, fixing this assertion error in one of the unit tests on one of the >>> buildbots: >>> >>> >>> http://fenicsproject.org:8010/builders/dolfin-master-full-wheezy-amd64/builds/1707/steps/make%20run_unittests_py/logs/stdio >>> >>> num_iter = solver.solve(x, b) >>> > assert num_iter == num_iter_ref >>> E assert 27 == 6 >>> >>> Looks like a very peculiar test to me. >>> >>> >>> >> What's peculiar about it? >> >> The problem is almost certainly an old version of PETSc on the buildbot. >> The test could be disabled for old versions of PETSc. This might request >> adding a function that returns the PETSc version. >> >> Garth >> >>> >>> mån 27 juli 2015 kl 11:53 skrev Jan Blechta <blec...@karlin.mff.cuni.cz >>> >: >>> >>>> I've got rid of the last forgotten issue targeted to 1.6. Are we >>>> waiting for something else? >>>> >>>> Jan >>>> >>>> >>>> On Fri, 17 Jul 2015 21:26:33 +0100 >>>> "Garth N. Wells" <gn...@cam.ac.uk> wrote: >>>> >>>> > I think we're ready now for the 1.6 release. >>>> > >>>> > Garth >>>> > >>>> > On 9 July 2015 at 22:54, Cian Wilson <cwil...@ldeo.columbia.edu> >>>> > wrote: >>>> > >>>> > > Done. See: >>>> > > >>>> https://bitbucket.org/tferma/dolfin/branch/systemassemblerfix-master >>>> > > >>>> > > Cheers, >>>> > > Cian >>>> > > >>>> > > >>>> > > On 07/08/2015 04:03 AM, Garth N. Wells wrote: >>>> > > >>>> > > >>>> > > On 25 June 2015 at 21:20, Cian Wilson <cwil...@ldeo.columbia.edu> >>>> > > wrote: >>>> > > >>>> > >> I still have a branch that fixes that and have been actively using >>>> > >> it since 1.5.0 came out. >>>> > >> >>>> > >> Happy to bring it up to current master and take any advice on >>>> > >> cleaner fixes if that would help. >>>> > >> >>>> > >> >>>> > > It would be a big help if you could bring your branch up-to-date >>>> > > with master. Thanks. >>>> > > >>>> > > Garth >>>> > > >>>> > > >>>> > > >>>> > >> Cheers, >>>> > >> Cian >>>> > >> >>>> > >> >>>> > >> On 06/25/2015 04:14 PM, Garth N. Wells wrote: >>>> > >> >>>> > >>> On 25 June 2015 at 19:36, Anders Logg <l...@chalmers.se> wrote: >>>> > >>> >>>> > >>>> What is the status of the upcoming release? Which are the current >>>> > >>>> blockers? >>>> > >>>> >>>> > >>>> This one is a blocker: >>>> > >>> >>>> > >>> https://bitbucket.org/fenics-project/dolfin/issue/494 >>>> > >>> >>>> > >>> Garth >>>> > >>> >>>> > >>> -- >>>> > >>>> Anders >>>> > >>>> >>>> > >>>> >>>> > >>>> _______________________________________________ >>>> > >>>> fenics mailing list >>>> > >>>> fenics@fenicsproject.org >>>> > >>>> http://fenicsproject.org/mailman/listinfo/fenics >>>> > >>>> >>>> > >>>> _______________________________________________ >>>> > >>> fenics mailing list >>>> > >>> fenics@fenicsproject.org >>>> > >>> http://fenicsproject.org/mailman/listinfo/fenics >>>> > >>> >>>> > >> >>>> > >> _______________________________________________ >>>> > >> fenics mailing list >>>> > >> fenics@fenicsproject.org >>>> > >> http://fenicsproject.org/mailman/listinfo/fenics >>>> > >> >>>> > > >>>> > > >>>> > > >>>> >>>> _______________________________________________ >>>> fenics mailing list >>>> fenics@fenicsproject.org >>>> http://fenicsproject.org/mailman/listinfo/fenics >>>> >>>
_______________________________________________ fenics mailing list fenics@fenicsproject.org http://fenicsproject.org/mailman/listinfo/fenics
