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: next round of 1.9.0 backports #16420
MAINT: next round of 1.9.0 backports #16420
Conversation
* apply the patch Matt suggested in scipygh-16337 * I was easily able to reproduce the failure reported there on a Linux box with a Python 3.9 venv where NumPy `1.19.3` was used at test/runtime to mimic the wheels repo, and confirm that the tests pass with this patch
* suppress rather than check for a warning in `test_theilslopes_warnings()` because the warning does not show up for 32-bit Windows with at least some versions of NumPy (issue for wheels repo)
* this is effectively a forward port and modernization of the release branch `PROPACK` shims that were added in scipygh-15432; in short, `PROPACK` + Windows + some linalg backends was causing all sorts of trouble, and this has never been resolved * I've switched to `SCIPY_USE_PROPACK` instead of `USE_PROPACK` for the opt-in, since this was requested, though the change between release branches may cause a little confusion (another release note adjustment to add maybe) * I think the issues are painful to reproduce; for my part, I did the following just to check the proper skipping/accounting of tests: - `SCIPY_USE_PROPACK=1 python dev.py -j 20 -t scipy/sparse/linalg` - `932 passed, 172 skipped, 8 xfailed in 115.57s (0:01:55)` - `python dev.py -j 20 -t scipy/sparse/linalg` - `787 passed, 317 skipped, 8 xfailed in 114.80s (0:01:54)` * why am I doing this now? well, to be frank the process of manually backporting this for each release is error-prone, and may cause additional confusion/debate, which I'd like to avoid. Besides, if it is broken in `main` we may as well have the shims there as well. I would point out that you may want to add `SCIPY_USE_PROPACK` to 1 or 2 jobs in CI? The other reason is that if usage of `PROPACK` spreads, I don't want to be manually applying more skips/shims on each release (which I already had to do here with two new tests it seems)
* update the error message to reflect the new environment variable given reviewer feedback (`USE_PROPACK` -> `SCIPY_USE_PROPACK`)
* set `SCIPY_USE_PROPACK=1` for meson jobs on all three platforms in CI * fix linter issues...
Closes scipygh-15296 (now really)
* ensure that two HiGHS-related tests that segfault for 32-bit Linux runs in the wheels repo are not only marked with `xfail`, but do not execute at all (to avoid segfaulting, of course) * for one of the tests, this will be conditional on 32-bit platform * for the other test, there was already an `xfail` mark so I've just bumped it up to not execute for now
* update author/issue/PR lists * mention the variable rename to `SCIPY_USE_PROPACK`
@@ -13,7 +13,7 @@ requires = [ | |||
"meson-python>=0.5.0", | |||
"Cython>=0.29.21", | |||
"pybind11>=2.4.3", | |||
"pythran>=0.9.12", | |||
"pythran>=0.11.0", |
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.
pretty sure this will create a merge conflict, but that's all being dealt with in gh-16353 anyway, where the caps are added.
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 doc build failure is:
WARNING: unsupported theme option 'logo_link' given
Because of the new version of the theme - can be fixed by including gh-16374.
The pre-release job shows a bunch of issues that still need fixing in main
I think, so can be ignored.
The rest LGTM
Ok, I cherry-picked Pamphile's theme changes and updated the relnotes again. Note that I did some of this manually because I keep hitting the API rate limit on github requests because everything gets routed through a proxy at work and the |
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.
LGTM now, thanks @tylerjereddy
Re gh_lists.py
, the issue is that we're still using the REST API and it should switch to the GraphQL API at some point to avoid issues with having to send many requests.
SCIPY_USE_PROPACK
env var renamepython -m pip install -v .
passes in avenv
with thisbranch, and the first 10 % of the testsuite, but that's about all I've checked here
so far
isn't mixed in with the finer details of the version pins and
meson
refinement