Thanks for the PR! I'm just marking it as "2 - In progress" while we wait for CI.

FWIW — hoping to be helpful — I found it not super easy to use the repo — I notice it doesn't use pytest / code formatters / etc. That does make contributions a bit harder for newcomers, though I'm sure it's for good reason...

Thanks for the comments, I do think they're helpful. It doesn't use pytest because there's a lot of homegrown addons to unittest that have been added over the years, and porting them over to pytest wasn't trivial (there was an attempt some years ago but I believe it stalled).

WRT to code formatting, we only check for flake8 compliance (with a large number of files excluded as they were historically never checked and they never got fixed up yet). I think there's also more that could be done to make things a smoother experience here, but it's been difficult to move forward with relative to other items.

From a quick skim it looks like the tests aren't platform-agnostic WRT the way paths are formed (the root, and separators) - I haven't looked deeper, but perhaps pathlib.Path could be used to make the form of the tests platform-independent?

From a quick skim it looks like the tests aren't platform-agnostic WRT the way paths are formed (the root, and separators) - I haven't looked deeper, but perhaps pathlib.Path could be used to make the form of the tests platform-independent?

Thanks, fixed

It doesn't use pytest because there's a lot of homegrown addons to unittest that have been added over the years, and porting them over to pytest wasn't trivial (there was an attempt some years ago but I believe it stalled).

Yes totally re homegrown addons...

I would say that pytest is much much nicer, both from the perspective of maintainers being able to design better test suites, and the perspective of contributors who are running tests — for example I couldn't seem to enter pdb on a test failure with the current setup? Transitioning probably isn't trivial but also probably not herculean — pytest can collect & run standard unittests, and running two test commands for a while could be a way to get moving.

WRT to code formatting, we only check for flake8 compliance (with a large number of files excluded as they were historically never checked and they never got fixed up yet). I think there's also more that could be done to make things a smoother experience here, but it's been difficult to move forward with relative to other items.

OK, you might want to check out ruff for linting; it fixes most failures itself. For formatting: we moved to black in a big bang in xarray, was very successful. The one-off merge conflicts are a bit annoying, but tractable ("merge up to the formatting commit, format with black, merge in the formatting commit").

The pytest transition probably requires some expertise but if you wanted a hand with linting & formatting, happy to help.

This is ready for review!

This is ready for review!

Oops, sorry, I forgot to check back when I was waiting for CI the other day.

Hi team — the gentlest of pings here. I know things can get backlogged. I do think this is a fairly small unambiguous improvement, with some decent comments and a great test. If there's anything I can do to reduce the reviewing burden, please lmk.

Sorry! I'll review it today

Sorry, I wrongly pushed to your branch an empty commit

No prob re empty commit, push away.

Docs error seems unrelated though?

numba/_helperlib.c:22:44: error: ‘I’ undeclared (first use in this function)
   22 |     #define _complex_float_ctor(r, i) (r   I * i)
      |                                            ^
numba/_helperlib.c:147:12: note: in expansion of macro ‘_complex_float_ctor’
  147 |     *out = _complex_float_ctor((float) _out.real, (float) _out.imag);
      |            ^~~~~~~~~~~~~~~~~~~
numba/_helperlib.c:22:44: note: each undeclared identifier is reported only once for each function it appears in
   22 |     #define _complex_float_ctor(r, i) (r   I * i)
      |                                            ^
numba/_helperlib.c:147:12: note: in expansion of macro ‘_complex_float_ctor’
  147 |     *out = _complex_float_ctor((float) _out.real, (float) _out.imag);
      |            ^~~~~~~~~~~~~~~~~~~
error: command '/usr/bin/gcc' failed with exit code 1

Docs error seems unrelated though?

Yes! I guess something must have changed on conda-forge.

I think that fixing the docs issue requires merging numba:main into this branch so that #9675 is picked up.

(I moved the dev tools discussions to other issues. This PR is ready for a second review — thanks!)

@sklam, are you ok with the current state or would like to review this one as well? If not, we can merge it.

@max-sixty, can you add a news fragment for this PR? It will then be part of the release note. See https://github.com/numba/numba/tree/main/docs/upcoming_changes for README and examples.

Great, added!

/azp run

Done!

No urgency but it states still Waiting on Author — I think it's now ready for you to merge — lmk if that's not correct