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
MAINT: additional 1.8.1 backports/adjustments #16146
MAINT: additional 1.8.1 backports/adjustments #16146
Conversation
…ces. test for using dot with a scalar quantity FIX: allow scalar multiplication when using dot() on sparse matrices
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.
The backported PRs all look reasonable. There are a lot of failures though. The ones due to a distutils
deprecation warning can be ignored. These ones may be due to the UNU.RAN update (not sure, @tirthasheshpatel do you remember?)
RuntimeWarning: invalid value encountered in multiply
No, I don't think the failures are related to UNU.RAN. The runtime warnings might be because of a change in NumPy. The prerelease_deps_coverage_64bit_blas uses the latest NumPy (1.23.0dev0). |
For the failures @rgommers is worried about, isn't that just the thing that @seberg was shimming for bleeding-edge NumPy in gh-16152? It looks like he is even working on the same file over there I also see a comment that the exact nature of the warnings is still subject to change, but I don't mind if we want to try backporting that when it is ready before we proceed here? |
The scalar math in NumPy gives a slightly different warning then the ufunc machinery. In this particular case, the ufunc machinery is now used, rather than the array case. That might change again (the ufunc path is quite a bit slower), but in general the warning given by the ufunc path is also much more useful so just fortify the tests. Note, one test failure remains due to: numpy/numpy#21481
Ok, I dealt with the merge conflicts to pull in the new-NumPy shims from Sebastian and updated the relnotes. Still, there are more failures to scan over I think, hopefully this reduces the count a bit more though. |
The prerelease job is timing out now, making it harder to assess remaining failures--I'll try a restart or two. |
For the NumPy prerelease failures, ignoring the distutils deprecation errors, we still have failures in:
@rkern @tupui @tirthasheshpatel looks like there may be some disagreement on the licensing backports (gh-16155, gh-16158)--suggestions? |
I believe that this commit should be reverted, yes. The rest of the PR is fine. |
I will make the fix for the licensing. But fine with me to not block the release for that. |
If you need it to fix test failures in the 1.8.1 release, then sure, backporting is fine. It is just a one-line fix of a test (three if you include the comments), so it should be straightforward to include it. |
@tylerjereddy gh-16163 is in. I believe this is good now. |
Thanks! Yeah, main delay is just that I'm still recovering from some kind of virus, hopefully I can resume here soon. |
oh no worries, take care of yourself first! |
…st equal'. The constant upper bound of the dB gain in the stop band is increased to -150*(1-1e-12), effectively converting the test into 'less or almost equal'.
MAINT: better/corrected UNU.RAN licensing information
This change can be reverted once Pip releases its next version with a fix for pypa/pip#11116. At the moment all our Azure builds are failing with errors like: ``` ERROR: Some build dependencies for file:///D:/a/1/s conflict with the backend dependencies: numpy==1.21.4 is incompatible with numpy==1.18.5; python_version=='3.8' and (platform_machine!='arm64' or platform_system!='Darwin') and platform_machine!='aarch64' and platform_python_implementation != 'PyPy'. ``` [skip github]
…#16175) Closes scipygh-16174. Note that this will likely be fixed in NumPy, but the test is fine written this way and it fixes an annoying CI failure, so we can leave the test like this. [skip azurepipelines]
I've updated to backport everything I think I need here, along with another release notes update. I'll wait for CI because we had so many failures before. My current plan is to allow for the following types of test suite failures with pre-release NumPy:
So, if I just see those failures I'll be tempted to self-merge. If I see additional failures with pre-release NumPy I may have to check those again. |
There are only two failing CI jobs left:
|
.dot
method of sparse matrices #16035git
security measure