Numpy 2.0 for packages that aren't pinned to numpy version?

Package Builders conda-forge
1

I’ve run into a fair number of packages that don’t work with numpy 2.0.

some crash out, which is bad

At least one “…won’t crash, but the result is nonsense.”

which is really bad.

Many of these have an unpinned “numpy” as a dependency.

I can add a <2.0 to the deps in the recipe, but the already built packages , of course, son’t have that, so presumably the resolver would pull an older version (older build?) if numpy is >=2.

How do we deal with this ?!?!

NOTE: the problem at hand is:

which is a complied extension. As it was compiled against numpy, will it, at least not try to run with numpy 2.0? ( I haven’t tested it yet…)

2

You will need to add upper bound pins in the repodata patches, similar to scanpy v1.10.2 requires numpy<2 by daylinmorgan · Pull Request #792 · conda-forge/conda-forge-repodata-patches-feedstock · GitHub.

3

Thanks – I’ll give that a try.

4

hmm – what is:

timestamp_lt: 1720623323000

(That value from the scanpy example)

Never mind, found it in the nested README :slight_smile: