REL: Prepare for NumPy 1.21.6 release.

[charris]

• Create 1.21.6-changelog.rst.
• Update 1.21.6-notes.rst.
• Fix circleci doc builds.
• Change azure-pipelines Mac vmImage to macOS-1015.

[seberg]

Sorry, I hadn't realize this was coming now. Can you hold off until tomorrow?

[charris]

Can you hold off until tomorrow?

Sure. This came up at the end of scipy/oldest-supported-numpy#45. Do you intend to add something?

[seberg]

Yeah, there is one small fix I have been meaning to do for a while and it should be quick.

[seberg]

Oh, just realized that this is 1.21.x not 1.22.x. Still happy to backport, though, but maybe not as pressing. (and there is a tiny chance of backcompat issues, so only if a 1.21.7 wouldn't be too much hassle.)

[charris]

I'd rather avoid any potential problems. I wouldn't do this at all if the oldest-supported-numpy problem hadn't come up.

[charris]

A 1.22 fix would be welcome, though. Things there have quieted a lot since 1.22.3, so probably only one more release.

[Czaki]

if this release will be able to build from source on Windows? CCompiler_spawn() got an unexpected keyword argument 'env' observed when building numpy==1.21.5 from source using latest setuptools:

          File "C:\Users\RUNNER~1\AppData\Local\Temp\pip-build-env-v9k16mu4\overlay\site-packages\setuptools\_distutils\_msvccompiler.py", line 505, in spawn
                  return super().spawn(cmd, env=env)
          File "C:\Users\RUNNER~1\AppData\Local\Temp\pip-install-u4z7ehkd\numpy_abda04e05c5e4cc9b598abfa5c2eb061\numpy\distutils\ccompiler.py", line 90, in <lambda>
                  m = lambda self, *args, **kw: func(self, *args, **kw)
        TypeError: CCompiler_spawn() got an unexpected keyword argument 'env'

https://github.com/4DNucleome/PartSegCore-compiled-backend/runs/5972857498?check_suite_focus=true#step:4:708

As I read test-related scripts the test process uses the system python which does not use latest setuptools.

[charris]

@Czaki NumPy often does not build with latest setuptools, which are changing regularly these days. We are pinned to older versions and will change the build to use meson at some point on account of the instability. Windows builds are difficult in any case and are best avoided when possible.

[charris]

@BvB93 I'm trying to do a 1.21.6 release, but am now getting typing errors for no reason I can see. Could you take a look at the error messages and see if there is anything that can be easily fixed? Something has changed.

EDIT: The tests run without problems on my machine, the errors seem to be Mac specific, and possibly due to the azure-pipeline environment.

[Czaki]

@charris Thanks for your response. Now I check and setuptools is pinned in pyproject.toml. So I do not understand why it fails.

I would love to avoid numpy builds but I also would love to deliver pypy wheels.

[mattip]

@Czaki you can get PyPy wheels for windows from the weekly builds

pypy -m pip install -i https://pypi.anaconda.org/scipy-wheels-nightly/simple numpy==1.23.0.dev0

If this doesn't work or is not what you want, please open a separate issue.

[BvB93]

EDIT: The tests run without problems on my machine, the errors seem to be Mac specific, and possibly due to the azure-pipeline environment.

Same here on my Mac; locally the tests pass without issue. It's not clear to me what could be causing the issue here, though based on the reproducibility issue presumably something related to azure?

[charris]

presumably something related to azure?

Sounds like it, they have broken things before, it is really annoying. Guess I'll put this in and see if the wheel builds pass. If not, is there an easy way to disable the tests?

[BvB93]

Sounds like it, they have broken things before, it is really annoying. Guess I'll put this in and see if the wheel builds pass. If not, is there an easy way to disable the tests?

We could always add an xfail for Mac with these specific tests.

[charris]

We could always add an xfail for Mac with these specific tests.

The wheel builds passed. Not sure what the differences are, need to take a look at that someday.

[BvB93]

Sure, we can always add an xfail for Mac if worst comes to worst. On 12 Apr 2022, at 01:24, Charles Harris ***@***.***> wrote:  presumably something related to azure? Sounds like it, they have broken things before, it is really annoying. Guess I'll put this in and see if the wheel builds pass. If not, is there an easy way to disable the tests? — Reply to this email directly, view it on GitHub<#21323 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AKK4FQ6SUIGQJUNUPUWSYJLVESYCVANCNFSM5TDVTGZA>. You are receiving this because you were mentioned.Message ID: ***@***.***>