Add pypi release and pipeline

[mpiannucci]

Add pypi project and create CI/CD pipelines for distributing to pypi

[ChrisBarker-NOAA]

Great idea -- but probably need to get it settled down to an initial release first.

Also -- once ready -- add it to conda-forge as well.

[ChrisBarker-NOAA]

NOTE: we may want to put linting into the CI first.

ruff?

-CHB

[ocefpaf]

Folks, for the sake of reserving the name xarray-subset-grid maybe we should consider a dev release? I can do that and setup the CI if you are OK with it.

[ChrisBarker-NOAA]

Yes! I was hoping to get to that in the next couple days....

I'd like to do a 0.1 rather than dev though.

NOTE: we can't do that until we resolve some test failures with the latest zarr - which may be as simple as pinning zarr.

[ocefpaf]

NOTE: we can't do that until we resolve some test failures with the latest zarr - which may be as simple as pinning zarr.

I believe that this PR will fix it: zarr-developers/zarr-python#2774

However, if I understand the error correctly, the failure is when fetching the data for the tests, not xarray-subset-grid code itself, right? I'd argue for a small local file if you want an all green CI, and/or a known_failure skip on that test.

[ChrisBarker-NOAA]

I believe that this PR will fix it: zarr-developers/zarr-python#2774

that would be nice -- though not sure when it would get released, so for now we need another path.

(though the pending fix makes me think pinning is a fine option)

However, if I understand the error correctly, the failure is when fetching the data for the tests, not xarray-subset-grid code itself, right?

I haven't dug into it yet, but probably, yes - the core code works with a xarray Dataset, and it should be agnostic to where / how it's backed.

However -- the impetuous for this project to be able to efficiently access kerchunked data on AWS. So we should probably have that somewhere in the tests, and certainly in the examples, which should run out of the box.

I'd argue for a small local file if you want an all green CI,

That would be good, yes.

and/or a known_failure skip on that test.

not sure I love that. We do have an --online flag for the tests, so that ones that need to be online will only run if you pass that flag -- but we do want them to work :-)

My vote at the moment -- is one of:

A) Find another way to get the tests to work with latest zarr -- do we need to use FSMap, or there another way to do it?

B) pin zarr<3 -- if only for the tests.

[ChrisBarker-NOAA]

NOTE: with regard to:

I'd argue for a small local file if you want an all green CI,

The tests need a lot of help, specifically: #13

and coverage is pretty bad.

So:
a) any help you can offer would be great!
and/or
b) whatever we can do to get a release out is fine

[ChrisBarker-NOAA]

Hmm:

https://pypi.org/project/zarr/#history

seeing all those yanked releases makes me think pining to zarr<3 for now is a fine idea.

[ChrisBarker-NOAA]

One more thing and we re done:

conda-forge -- with a PR ready to be approved:

conda-forge/staged-recipes#30391

And then we are fully published!