New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Use external ci-helpers #392
Conversation
@bsipocz - Thank you! There's two builds where Cython is missing: |
It's just a side effect of the package version conflicts I've mentioned
|
Now we're getting this error in Gammapy on a different PR (#396):
https://travis-ci.org/gammapy/gammapy/jobs/96029239#L1210 @mwcraig - Did something change in conda or the Astropy channel builds? |
I can look at it tomorrow, and will try to fix it here. I'll let you know
|
Took a quick look (it was that or grade :)). One of your requirements is pulling in numpy 1.8 (sherpa), others are using 1.9 and others 1.10. conda does not update already-installed packages if you change the version of numpy installed. The easiest way to ensure the version you want is to include |
@mwcraig - Thanks for having a look and advising! I think we don't really need to pin the numpy version. Astropy and Sherpa should work with numpy 1.8 or 1.9 or 1.10, no? I think it's at least worth a try to "unpin" Numpy in some of the builds and see if they pass, i.e. the different Numpy versions are compatible enough for our needs. @bsipocz - Let me know if you don't get to it tomorrow or get stuck. I have some time on Saturday and can try to help finish up this PR and fix the new numpy version issue. |
@cdeil @bsipocz -- sorry, I think I wasn't very clear. Some of your dependencies already have their numpy dependencies pinned. For example, sherpa is built against numpy 1.8 (its build string is The problem is that the first set of numpy dependent packages currently being installed is compiled against numpy 1.9. The next set probably just asks for the most recent numpy (or maybe astropy) and gets packages built agains numpy 1.10 and numpy 1.10 instead of 1.9. Finally, sherpa gets installed, and changes numpy to 1.8. So in the end you have a mix of binary packages, some built against numpy 1.9, some against 1.10, and one against 1.8, and the final version of numpy installed is 1.8. You might be able to fix this just by reinstalling numpy 1.10 at the very end since both the numpy ABI and API are supposed to be forward-compatible. The alternative is to pin numpy everywhere to make sure everything gets installed with the same numpy. |
624e3d2
to
3dd027e
Compare
@bsipocz - Currently this PR fails on travis-ci like this: I can't reproduce the issue locally, but I think it's something that was fixed in @bsipocz - Could you please bump the scipy version number required and used on travis-ci to 0.15 and see if that resolves the issue? |
@cdeil - There is no scipy >=0.15 build available in conda with np1.8. Would you prefer to have a build against np1.8 with Option a) seems to be good enough, but I don't know much about gammapy, so I may miss something. |
I'm not sure what your options imply for our test coverage, i.e. what the final build matrix would look like. The most important build for us is one that does include scipy and sherpa, because we use scipy for interpolation and sherpa for fitting in our main analysis routines. If none of the builds contains both scipy and sherpa, that's not good. (we had this before, so it should be possible, even if tedious with having to pin numpy version or run install commands in the right order). |
OK, I'll rebase then to get rid of the noise commit here, and put back the full matrix. |
…pers (didn't manage to work out how to get it installed without conflicts (typically with naima)"
134be5f
to
f5be277
Compare
f1d5639
to
12a3a1c
Compare
12a3a1c
to
f8cfffa
Compare
d3d85ee
to
76dc28e
Compare
76dc28e
to
8a4c2f3
Compare
…ne triggered by upstream sherpa that we can't do anything about
@cdeil - I've disabled enable_deprecations_as_exceptions() as an exception was raised in sherpa that I couldn't get rid of. It can be turned back on once the astropy issue is solved astropy/astropy#4430 |
@cdeil - This is ready for review. |
@@ -7,6 +7,8 @@ | |||
from astropy.coordinates import Angle | |||
from astropy.io import fits | |||
from astropy.wcs import WCS | |||
# Remove this when/if https://github.com/astropy/astropy/issues/4429 is fixed |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Please add TODO:
here so that we can find this later and clean it up.
@bsipocz - Thank you! Happy new year! I mentioned two minor things inline. Otherwise this looks good to merge to me! There's no reason to keep the bundled |
@cdeil - Happy new year to you, too! I've added the |
@bsipocz - Thank you! I know that figuring out build / test issues as you did in this PR is difficult and time consuming. |
Unfortunately I couldn't manage to pass the sherpa build to
ci-helpers
as it always ended in conflicts just as @cdeil predicted.As a working solution it gets installed after the
ci-helpers
script finishes, directly from.travis.yml
.It still has issues with the older numpy-s, I'll come back to it.This PR also updates to use np 1.10 and astropy 1.1, and adds extra tests for np 1.9 and astropy lts.
Closes #391