Add automated windows CI using appveyor#2836
Conversation
Got a build fail with the previous version....
[skip ci]
...because the prior commit turned skipped ci to avoid an unnecessary run on appveyor.
|
I have added the astropy, astropy-helpers, and package-template repositories to the appveyor astropy account. |
|
If we go ahead with this solution, can we make the It looks like this includes both 32- and 64-bit builds. @Cadair, does yours include that too? What are the main differences between the two solutions? |
|
I can add 32bit builds with one extra line in the config file. I spent some time on mine getting rid of some of the setup code that isn't needed when using miniconda, which still seems to be in this version. |
|
This version also does a little more in the powershell install script, where as #2755 does mostly everything in the Otherwise I don't think there is a lot of difference, they are both based on the same code. |
|
Mine also has the miniconda version set from the It seems that @mwcraig's handle on powershell is better than mine so he did more stuff in the powershell where as I did as much as I could get away with in the yml. |
|
😊 Can't believe I didn't check the list of open issues first... In any event, this version compiles on 64-bit -- the 64-bit fails here are real test fails, but I think I know which piece in mine would fix @Cadair's. To keep things simpler, I'll close this PR and make a few PRs on @Cadair's fork instead. There will be at least two files beyond Just to be clear I don't know much of anything about powershell, but I'm really good at copy and paste and happened upon a more recent variant of this than you had :) |
|
@mwcraig as far as I can tell we don't need |
|
@Cadair -- I think we do actually need it. The difference between two 64-bit builds I tried is below. The first, which did not set the environment variables, failed at compile (https://ci.appveyor.com/project/mwcraig/astropy/build/1.0.4#L915) The second, which set the environment variables, succeeded in compiling, though three tests failed (https://ci.appveyor.com/project/mwcraig/astropy/build/1.0.7#L2073) My limited experience with compiling 64 bit in windows for conda is that magic is usually required, unfortunately... diff --git a/appveyor.yml b/appveyor.yml
index 456c247..22eef85 100644
--- a/appveyor.yml
+++ b/appveyor.yml
@@ -44,7 +44,7 @@ test_script:
# doesn't really work.
#
#- "%CMD_IN_ENV% conda build conda-recipe --quiet"
- - python setup.py test
+ - "%CMD_IN_ENV% python setup.py test"
#artifacts:
# Archive the generated conda package in the ci.appveyor.com build report. |
…onger needs to include its own copy of the make_func_with_sig utility. I also renamed make_func_with_sig to make_function_with_signature as suggested here: astropy#2835 (comment) Previously I thought the unabbreviated name was a little on the long side. But on second thought I prefer the clarity here.
…onger needs to include its own copy of the make_func_with_sig utility. I also renamed make_func_with_sig to make_function_with_signature as suggested here: astropy#2835 (comment) Previously I thought the unabbreviated name was a little on the long side. But on second thought I prefer the clarity here.
…onger needs to include its own copy of the make_func_with_sig utility. I also renamed make_func_with_sig to make_function_with_signature as suggested here: astropy#2835 (comment) Previously I thought the unabbreviated name was a little on the long side. But on second thought I prefer the clarity here.
…onger needs to include its own copy of the make_func_with_sig utility. I also renamed make_func_with_sig to make_function_with_signature as suggested here: astropy#2835 (comment) Previously I thought the unabbreviated name was a little on the long side. But on second thought I prefer the clarity here.
…onger needs to include its own copy of the make_func_with_sig utility. I also renamed make_func_with_sig to make_function_with_signature as suggested here: astropy#2835 (comment) Previously I thought the unabbreviated name was a little on the long side. But on second thought I prefer the clarity here.
…onger needs to include its own copy of the make_func_with_sig utility. I also renamed make_func_with_sig to make_function_with_signature as suggested here: astropy#2835 (comment) Previously I thought the unabbreviated name was a little on the long side. But on second thought I prefer the clarity here.
…onger needs to include its own copy of the make_func_with_sig utility. I also renamed make_func_with_sig to make_function_with_signature as suggested here: astropy#2835 (comment) Previously I thought the unabbreviated name was a little on the long side. But on second thought I prefer the clarity here.
…onger needs to include its own copy of the make_func_with_sig utility. I also renamed make_func_with_sig to make_function_with_signature as suggested here: astropy#2835 (comment) Previously I thought the unabbreviated name was a little on the long side. But on second thought I prefer the clarity here.
…onger needs to include its own copy of the make_func_with_sig utility. I also renamed make_func_with_sig to make_function_with_signature as suggested here: astropy#2835 (comment) Previously I thought the unabbreviated name was a little on the long side. But on second thought I prefer the clarity here.
|
Good news everyone! Looks like GitHub now officially supports having multiple CI statuses :) https://github.com/blog/1935-see-results-from-all-pull-request-status-checks |
|
woop! woop! |
|
Nice! |
|
I wonder why I referenced this PR in so many commits... must have been a typo. Sorry 'bout that. |
This PR adds automatic testing on windows using appveyor.com.
It is based on, and incorporates parts of, https://github.com/rmcgibbo/python-appveyor-conda-example by @rmcgibbo, which is in turn a fork of https://github.com/ogrisel/python-appveyor-demo by @ogrisel. Honestly, they did all of the hard work!
To see an example of the report produced, go to: https://ci.appveyor.com/project/mwcraig/astropy/build/1.0.8
It does not test the same range of python/numpy variants as the travis tests, but does cover the basics: python 2.7 and 3.4 on both 32 and 64 bit.
I took a quick look at the failures in 64-bit...they look like real test failures rather than problems with the appveyor configuration.
Before this is merged we would need to:
I hope to soon make a couple of additional pull request mining the goodness in the appveyor examples to automatically build conda packages of the dev version of astropy (for travis) and for affiliated packages (for linux and windows)....but this was the most straightforward place to start.