Remove NumPy <2 pin, remove panel ceiling

[seberg]

This PR removes the NumPy<2 pin which is expected to work for
RAPIDS projects once CuPy 13.3.0 is released (CuPy 13.2.0 had
some issues preventing the use with NumPy 2).

EDIT: Sorry, converted back to draft again, NumPy 2 is not yet picked up.

[jakirkham]

Note that cuxfilter depends on cuspatial

cuxfilter/dependencies.yaml

Line 200 in ca63e9d

- &cuspatial_unsuffixed cuspatial==24.10.*,>=0.0.0a0

So this depends on PR: rapidsai/cuspatial#1441

[jakirkham]

cc @jameslamb (for awareness)

[seberg]

Yeah, sorry, I unmarked draft too early, but I think once things are in place and NumPy is picked up we can make it no-draft again and it should be good.

[jakirkham]

All good. Was more looping in James as we had discussed the cuSpatial NumPy 2 PR earlier

[jameslamb]

I've taken over rapidsai/cuspatial#1441 and so am testing here too.

Got cuspatial to a state where it can be installed alongside Numpy (2.0.2): rapidsai/cuspatial#1441 (comment)

I pushed commits here to use the packages built from that PR's CI. Let's see if it'll be enough to loosen the pin here!

[jameslamb]

Putting an explicit >=2.0 floor here, I see all builds and tests passing!

  - numpy                                     1.26.4  py312heda63a1_0                      conda-forge             Cached
  + numpy                                      2.0.2  py312h58c1407_0                      conda-forge                8MB

(build link)

That's evidence that cuxfilter and its dependencies can be installed alongside NumPy 2.x.

Now that rapidsai/cuspatial#1441 is merged, I just pushed commits removing all the testing-only changes here and moving that pin back down to >=1.23,<3.0.0a0.

Hopefully we'll see everything pass here and use NumPy 2.x.

[jakirkham]

Interested to see how this goes

If we need to ignore a specific version, maybe we can do !=

Though if this is just the case of working around a past solver issue or limiting to a set of tested releases, then simply dropping seems reasonable

Maybe we can follow up with the cuxfilter team on which makes more sense

[jameslamb]

Yep for sure!

Note this pin is only for the testing environment.... it's already unpinned (>=1.0) at runtime for users.

• context for why it was pinned (something about compatibility with parallel pytest tests): refactor CUDA versions in dependencies.yaml #562 (comment)
• no ceiling on run dependency:

  cuxfilter/conda/recipes/cuxfilter/meta.yaml

  Line 42 in cd699b8

  - panel >=1.0

@AjayThorve could the ceiling on panel be removed?

I do see evidence that it helps conda find a solve with cuxftiler=24.10.* and numpy>=2.0.

  + bokeh                                      3.5.2  pyhd8ed1ab_0                         conda-forge                5MB
  + numpy                                      2.0.2  py312h58c1407_0                      conda-forge                8MB
  + panel                                      1.5.0  pyhd8ed1ab_0                         conda-forge               21MB

(build link)

[AjayThorve]

Yeah, AFAIK, there are no issues at runtime, so as long as the tests pass (seems like they do), this should be fine

[jameslamb]

Great, thank you!

[AjayThorve]

LGTM!

[AjayThorve]

Yeah, AFAIK, there are no issues at runtime, so as long as the tests pass (seems like they do), this should be fine

[jameslamb]

/merge

[jakirkham]

Thanks all! 🙏