Repository navigation
Conversation
|
I'm getting an error for Travis CI that says:
so there may be a problem with Travis CI at the moment. In a while we should trigger another Travis CI build so that this issue will hopefully get resolved. |
This might work?
The instructions on the Travis CI webpage to use "3.7-dev" as the Python version hasn't worked out. I'm not sure how to go about testing against Python 3.7-dev, so in the meantime I added a few lines for possibilities.
|
I think your first try was correct (Here's the Travis from that one) as Travis does find Python 3.7-dev: But then astropy's ci-helpers fails because it tries to fetch python3.7-dev from conda and ends up installing Python3.6.3: Further down So I would revert this back to There may some flag or something you can throw at |
|
One hacky idea might be to add a line before |
|
Digging into ci-helpers, I think this line is the culprit So if no I'm going to add some lines to This would be placed before |
Hack to get around `source ci-helpers/travis/setup_conda.sh` not finding correct version of Python when `3.7-dev` is installed via Travis.
Attempting to fix ci-helpers getting wrong version of Python when working with `3.7-dev`. This hack binds the `3.7-dev` version of Python to `$PYTHON_VERSION` which is searched for by ci-helpers.
|
Any idea what this rake business is all about? |
|
Looks like the last commit which didn't have this rake error was fb857af, I think... |
I think using the wrong kind of white space is causing the rake error
|
Looks like my hacky bash script isn't working... it doesn't even echo the python version back to Travis. Or is that just suppressed because it is stdout instead of stderr? |
|
Progress! The script seems to sort of work, although it needs to be parsed out to just give the version numbers with no spaces |
Had an error where the whole string from `python --version` output was being bound to the environment variable (e.g., "Python 3.6.3") whereas we just want the number.
|
So version number now gets correctly passed using the hack script, and 3.6 tests are building again, but conda doesn't like 3.7. Does anyone know of a way to force conda to fall back on the system installed version of python? |
Setting DEBUG flag to find out where in `setup_conda.sh` the installation fails.
|
We are now indeed failing here where conda tries to Python 3.7-dev is definitely in Travis's PATH, but it is behind miniconda's version of python Maybe we can force conda to use the 3.7-dev version by rearranging PATH and symlinks, like so? EDIT: we might be able to leverage |
|
TODO:
|
|
I wonder if we can figure out how to test against astropy-dev as well, which we tried to figure out how to do in #172. |
|
for astropy-dev with ci-helpers you can just do |
|
@Cadair Does that also test against python 3.7 (or whatever current dev version of Python is) or does it just test the dev version of astropy? |
|
I'm closing this because figuring out a way to test against the development version of Python 3.7 is probably not worth the effort since it's going to be released in ~2-3 months anyway. |
Python 3.7 is now in beta, and is likely to be released around this summer. We should make sure our 0.1.0 release works with the most recent development version of Python 3.7 since our 0.2.0 release will probably not occur until a few months after the Python 3.7 release. This pull request adds a line in
.travis.ymlto test against Python 3.7-dev.