-
Notifications
You must be signed in to change notification settings - Fork 193
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
0.11.0 - kraoñenn #1946
0.11.0 - kraoñenn #1946
Conversation
@rgommers, @jeanlaroche, @stefanv, @hroncok, @drew-parsons, @mgorny, @h-vetinari : this is the release candidate for a new pythran release, with as major change the move from xsimd 7 to xsimd 8. I plan to do the release by the end of the week, feel free to report bug (if any!) |
I'm happy to test this in conda-forge, but it would be a fair bit easier with an rc-tag rather than just a PR. Would that be possible? |
All my builds are successful with 0.11.0 . Very happy about that! |
Using the branch name is jot an option ? |
Yes, that works. I hadn't seen that it's a public branch on the repo here. |
Tried this branch in conda-forge/pythran-feedstock#63 - PPC fails at test time with something related to xsimd (we're using 8.0.3 like vendored here, but coming from conda-forge), the other arches/OSes build & test fine.
|
cc @JohanMabille : I think it's fixed in current xsimd master branch, maybe we could do a minor release? |
I'm seeing that |
Basic build seems fine, testing on amd64 (debian unstable). test_complex_conj fails with xsimd 7.6, as you would expect (cf. #1943 ). All the tests that I ran (test_base.py) succeed with xsimd 8. Building the docs with sphinx gives this error: "module 'mistune' has no attribute 'BlockGrammar'"
But I can't see a reference to edit: looks like this doc problem is due to mistune, which was just updated to v2.0.0. cf. miyakogi/m2r#66 . Not a pythran error. |
There is a new mistune release 2.0.0 that breaks things. |
@hroncok what's "mistune" ? |
I don't know it, except that it's https://github.com/lepture/mistune . nbconvert has a filter that uses it. |
https://pypi.org/project/mistune/ is a dependency of nbconvert which is a dependency of nbsphinx. The |
Except they already did that years ago: jupyter/nbconvert@365683f No idea where is the error coming from then, sorry. |
Looks like the debian packaging for nbconvert is overriding or ignoring the |
8ef3eea
to
c11e1d8
Compare
I've updated the PR with the new version of xsimd. conda-forge has been updated as well and Fedora is being updated. This should fix most of the above issues (if not all!) |
Well, I'm getting a test failure on Python 3.10.1 (3.8 & 3.9 seem to pass 100%):
There's also a lot of warnings. Full build & test log (.xz, expands to 3.7M):
|
Happy to report no more failures on the conda-forge feedstock with xsimd 8.0.5 (note that as of now, the recipe is not running the full upstream test suite - that could be changed though) |
Thanks @mgorny. I suspect this is an unstable test. You can ignore it. |
Thanks @h-vetinari . I think I'll merge tomorrow then. |
I'll try today. Sorry for not getting to it sooner. |
There will be test failures on ppc64le as well as other architectures:
Full logs available only after it runs all tests. |
ppc64lehttps://koji.fedoraproject.org/koji/taskinfo?taskID=79788039
|
s390xhttps://koji.fedoraproject.org/koji/taskinfo?taskID=79788040
Seems like the same problem as on ppc64le. |
armv7hl will also fail. The remaining architectures (i686, x86_64 and aarch64) all passed. |
armv7hlhttps://koji.fedoraproject.org/koji/taskinfo?taskID=79788035
Again, this sounds very similar, if not the same, to the failures on ppc64le and s390x. |
A SciPy build with Pythran
With Pythran installed from source from this PR, plus xsimd I will do some testing on CI as well tonight to check Windows/macOS are happy. EDIT:
|
Yeah that's totally expected :-) |
c11e1d8
to
d808c9e
Compare
@hroncok I've pushed an extra patch that hopefully solves the |
I could also backport xtensor-stack/xsimd#655 to Fedora's xsimd package if that seems nicer. |
If it's applied in pythran source, we touch more than the Fedora users, so I don't think it's needed. But I did the patch without testing it, so I'd be happy with a green light from your side ;-) |
I thought I submitted a build yesterday evening, but I've made a mistake and the build did not see xsimd 8.0.5, hence retrying now. Sorry for the delay. |
ppc64le 1 test failure:
|
armv7hl:
|
I think we can ignore that one, it's at the rounding limit (the armv7 one) |
And the ppc one may be an LTO issue, I don't want to block the release for it. |
Thanks for all the reviews and tests. Tag 0.11.0 set, sources uploaded to PyPI |
Would you recommend disabling LTO on PPC or skipping the test? |
On Tue, Dec 14, 2021 at 02:58:58AM -0800, Miro Hrončok wrote:
And the ppc one may be an LTO issue, I don't want to block the release for
it.
Would you recommend disabling LTO on PPC or skipping the test?
The former
|
For the record, LTO has given problems also with pybind11 on mips64el and ppc64el, cf. https://lists.debian.org/debian-mips/2020/09/msg00000.html , https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1000780 |
Out of curiosity, how did the |
In Fedora, we skip (with pytest's
|
No description provided.