These are chat archives for astropy/astropy

15th
Jun 2016
Simon Conseil
@saimn
Jun 15 2016 08:21
The wiki page for RC testing (https://github.com/astropy/astropy/wiki/v1.2-RC-testing) supposes that tests are run without optional dependencies ?
And I get test failures when using optional dependencies, how should I report this ?
Brigitta Sipocz
@bsipocz
Jun 15 2016 12:23
@saimn - I think you should report any test failures. The standard CI can't cover all of the parameter space of possible systems setups, and the whole point of RC testing, as I see it, is to try it out on as many different systems as possible (especially that these are the one that our users have, not our dev ones).
Simon Conseil
@saimn
Jun 15 2016 12:27
@bsipocz - Yep, I have opened issues, but my question was more about the tables in the wiki page, and the fact that optional deps are not mentioned in this page. I guess I can just add a line in the "Failures" table, but it maybe useful to mention that it is on optional tests ?
Brigitta Sipocz
@bsipocz
Jun 15 2016 13:33
I guess you can have that as notes, or only say the basic parameters there as the other lines because the detailed setup is listed in the issues anyway.
Larry Bradley
@larrybradley
Jun 15 2016 14:05
travis-ci tests started failing last night for Astropy and affiliated packages (e.g. photutils, regions). Tests that passed as PRs subsequently failed when merged. The reason appears to be that travis-ci started using (an older?) numpy from astropy-ci-extras. cc: @bsipocz @mwcraig
Thomas Robitaille
@astrofrog
Jun 15 2016 14:05
@bsipocz - do you think we are running into conda/conda#2675
If so, we can just add defaults to the default channels?
Larry Bradley
@larrybradley
Jun 15 2016 14:11
@astrofrog That's probably it. I could find nothing that changed in ci-helpers or conda-channel-astropy last night. A change in conda's behavior would explain it.
Brigitta Sipocz
@bsipocz
Jun 15 2016 15:40
A fix in ci-helpers is merged, I've restarted the failing builds in the repos where I had travis rights. Someone should do the same for astropy.
Larry Bradley
@larrybradley
Jun 15 2016 16:04
@bsipocz The numpy 1.7 and 1.8 tests are still failing.
Larry Bradley
@larrybradley
Jun 15 2016 16:23
It's because conda is installing astropy 0.3.0(!) for those tests.
Brigitta Sipocz
@bsipocz
Jun 15 2016 16:28
Previously those test were passing as astropy 1.0.3 could have been picked up from the astropy-ci-extras channel.
So it wasn't very helpful either.
We agreed (based on the assumption that the channel is not really up-to date) that since it's hardy used, maybe it's better if the packages list it themself in their CONDA_CHANNELS variable.
However now I think that maybe we should add it back in, since now we see that all affiliates testing against old numpies picked up their astropy from there.
Larry Bradley
@larrybradley
Jun 15 2016 16:41
For the old numpy tests, can we explicitly add the -c astropy-ci-extras channel when installing just astropy, e.g. conda install -c astropy-ci-extras -n <env> astropy?
Brigitta Sipocz
@bsipocz
Jun 15 2016 16:41
@astrofrog - The issue is not with the version of conda/miniconda. As I see the problem comes from the fact that they don't build packages against older numpies, thus if we want to test against np 1.8 or 1.7 we are stuch with the astropy version that was new at the time.
One solution would be to add back astropy-ci-extras as a default channel, or use pip wheels?
yes, that's a good solution, though maybe a bit overcomplicate as we need to do it for all packages.
Michael Seifert
@MSeifert04
Jun 15 2016 16:43
Fortunatly I've tested using pip for astropy installations quite recently: https://github.com/astropy/package-template/issues/173#issuecomment-226059492 :disappointed:
Brigitta Sipocz
@bsipocz
Jun 15 2016 16:43
And still we will only get 1.0.3, as astropy-ci-extras is not maintained at the moment.
I'm thinking about adding something like this to ci-helpers: if ASTROPY_VERSION=stable it cannot install something that is older than the LTS, so the install process falls back to pip from conda.
Michael Seifert
@MSeifert04
Jun 15 2016 16:53
@bsipocz Would that work if there is another requirement that explicitly needs astropy?
Brigitta Sipocz
@bsipocz
Jun 15 2016 16:59
You have to give the astropy version somewhere even in that case, or if it is listed as their dependency then it would pick up something from conda just as it does now.
Brigitta Sipocz
@bsipocz
Jun 15 2016 17:53
ci-extras is back now, and I've restarted the regions and photutils builds
Thomas Robitaille
@astrofrog
Jun 15 2016 18:54
Thanks @bsipocz!
Larry Bradley
@larrybradley
Jun 15 2016 19:39
Thanks, @bsipocz! Everything is passing now.
Michael Seifert
@MSeifert04
Jun 15 2016 21:22
Is it normal that doctests tries to run code inside a .. plot::-directive and fail with RuntimeError: Invalid DISPLAY variable (the travis sphinx build works but all other tests fail). Is there a way around it?
Michael Seifert
@MSeifert04
Jun 15 2016 21:32
oh ... I should've read the .travis.yml more carefully. Please forget I asked that. :sweat_smile: