On Wed, Jul 05, 2017 at 09:12:55AM -0500, Roland Haas wrote:
Present: Ian, Roberto, Roland
I am sorry for missing the call - something local came up...
We are not going to hold the release if there are only 1-2 failures on machines (unless those failures are in core thorns).
I agree.
Some remarks:
- What machine is 105aff02? - Why are the Dissipation testsuites failing on BW/Comet/Edison/Cori/Golub/Hydra? (ADMConstraints? ... see below) On BW they don't seem to produce output at all, but on the others the same number of sufficiently different files suggest a common cause. - sphere_pugh_ppm is one of the 'fails a bit sometimes' thorns, and has been for a while. It's not a release-blocker, but a candidate we should look at. This might very well be a not optimal testsuite setup. - RotatingSymmetry90/180 testsuites also seem to fail on Golum/Edison/Cori. I seem to remember that, like Dissipation, they use ML_ADMConstraints results in the testsuite. Could you please check if the workaround is already applied in the source that is tested here, i.e., if the workaround does not work here? - We should probably disable OpenCL for BW if not even the WaveToy works there: keep it commented out in the optionlist, but don't enable it by default, because it might fool people in thinking it should work.
Overall, this looks really quite good - in fact, better than at some other times for a release.
Keep adding/updating testsuites, I'll prepare release notes and get everything ready for a release. For now, I will not create release branches before that status of ML_ADMConstraints is sorted out (patch is in the works, but not in master and not widely tested yet).
Frank