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
1.19.0rc2: test_full_reimport fails with "RuntimeError: implement_array_function method already has a docstring" #16508
Comments
I wonder why this isn't seen in the other testing environments, there has got to be some difference. |
What happens if you use the software versions in |
@seberg Thoughts? |
Thanks for the ping. This is weird because the error should read |
Hmm, or the test being called in a sub-process for some reason picks up a different NumPy version? And in that case, do we have to detect that? |
That's an interesting, but disturbing, thought :) |
Maybe running in a virtual environment would fix that? |
that's very likely what's happening! to build numpy, we install its build dependencies, some of which already depends on the packaged numpy, so that gets installed as well; f.e. from the build log in the original post:
so that's the system-wide available version of numpy. Then we start building the new version, and we do try hard to run tests/docs/examples against the being-built version, but running a subprocess with is |
@sethtroisi, yes we squeezed it into the 1.19 release, because it was annoying some users (even if the whole patch is at this time a bit dangerous). Hmmm, I am considering passing in the actual numpy version and skip the test when they do not match? |
Or wait. Are the environment variables not inherited here? So that e.g. |
You could try numpy's |
Ah, subprocess does inherit environment, so I suppose the issue is indeed the setup entirely running the previous NumPy? |
@sandrotosi (seberg) meant to ping you here |
the way we're running tests is by injecting the newly built location as the first element in |
@sandrotosi but maybe than we just should do that as well, I assume you mean |
it's i'm not sure that's a valid general approach for numpy upstream (f.e. what would happen if you install numpy and then just run |
Would instead running it using I thought maybe we can add the path of |
Or maybe you just can switch directories to anywhere else? |
I remember having a bit of a hard time figuring out the right way to run tests during the build process, since numpy tries really hard not to let you import it from the source dir :) we're already skipping a couple tests, thanks for your help on this, going to close this issue now and address it in debian! |
Blacklisting this one is definitely fine. I wish we could help you more though. I would think that simply change of directory (e.g. to a temporary one) would be fine, but I guess there are probably some subtleties I am missing... |
Fix for numpy#16508. Cython wants every pointer declaration to be on a separate line.
Hello,
this has started appearing with rc2 (maybe also rc1, but it didnt happen in 1.18.4):
this is running the test suite in our build env, which is a bare minimum chroot with only few packages installed; full log of an example is here: https://buildd.debian.org/status/fetch.php?pkg=numpy&arch=amd64&ver=1%3A1.19.0%7Erc2-1&stamp=1591237245&raw=0
similar issues (not sure how much relation there is): #14012, #14384, #15563
The text was updated successfully, but these errors were encountered: