[MNT] `conda` release - compatibility with Python 3.12

MCRE-BE commented 5 months ago

Desbribe the issue

Quick question, is scikit-base really compatible with Python 3.12 ? Or is this one of the issues you mention with scikit-base and conda releases being "under development" Under conda and mamba I cannot update to 3.12 (either because of SKTime, but likely because of the dependency on scikit-base).

(base) C:\WINDOWS\system32>conda create -n New python==3.12.3 scikit-base
Channels:
 - conda-forge
 - defaults
Platform: win-64
Collecting package metadata (repodata.json): done
Solving environment: - warning  libmamba Problem type not implemented SOLVER_RULE_STRICT_REPO_PRIORITY
warning  libmamba Problem type not implemented SOLVER_RULE_STRICT_REPO_PRIORITY
\ warning  libmamba Problem type not implemented SOLVER_RULE_STRICT_REPO_PRIORITY
warning  libmamba Problem type not implemented SOLVER_RULE_STRICT_REPO_PRIORITY
warning  libmamba Problem type not implemented SOLVER_RULE_STRICT_REPO_PRIORITY
warning  libmamba Problem type not implemented SOLVER_RULE_STRICT_REPO_PRIORITY
failed

LibMambaUnsatisfiableError: Encountered problems while solving:
  - package scikit-base-0.4.3-py310hbbfc1a7_0 requires python_abi 3.10.* *_cp310, but none of the providers can be installed

Could not solve for environment specs
The following packages are incompatible
├─ python 3.12.3  is installable with the potential options
│  ├─ python 3.12.3 would require
│  │  └─ python_abi 3.12.* *_cp312, which can be installed;
│  └─ python 3.12.3 conflicts with any installable versions previously reported;
└─ scikit-base is not installable because there are no viable options
   ├─ scikit-base [0.4.3|0.4.4|...|0.7.8] would require
   │  └─ python_abi 3.10.* *_cp310, which conflicts with any installable versions previously reported;
   ├─ scikit-base [0.4.3|0.4.4|...|0.7.8] would require
   │  └─ python_abi 3.11.* *_cp311, which conflicts with any installable versions previously reported;
   ├─ scikit-base [0.4.3|0.4.4|...|0.7.8] would require
   │  └─ python_abi 3.8.* *_cp38, which conflicts with any installable versions previously reported;
   └─ scikit-base [0.4.3|0.4.4|...|0.7.8] would require
      └─ python_abi 3.9.* *_cp39, which conflicts with any installable versions previously reported.

fkiraly commented 5 months ago

Thanks for reporting!

Indeed it seems this has not been properly updated in the conda recipe.

That explains the low download numbers, we have wondered why they are so much lower on conda.

With pip, there should be no problems though.

fkiraly commented 5 months ago

Attempted fix: https://github.com/conda-forge/skbase-suite-feedstock/pull/49

You wouldn't perhaps know what is going on with numpy there, in the CI jobs?

fkiraly commented 5 months ago

Something is going on here, here's a new PR on the sktime conda feedstock, this was not triggered by us - FYI @yarnabrina, what is this? https://github.com/conda-forge/sktime-feedstock/pull/123

fkiraly commented 5 months ago

Also summoning @dhirschfeld, @freddyaboulton - any idea what is going on here?

yarnabrina commented 5 months ago

Something is going on here, here's a new PR on the sktime conda feedstock, this was not triggered by us - FYI @yarnabrina, what is this? conda-forge/sktime-feedstock#123

I do not know. Can we ask help from conda forge maintainers who helped during previous release? I forgot their Github handle.

fkiraly commented 5 months ago

@beckermr, @mbargull, may I kindly ping you on this, since we seem to be out of our depth.

To summarize the history of this discussion:

@MCRE-BE noticed that the conda release of scikit-base does not seem to install for python 3.12 due to dependency conflict, that's the original post
however, we could find no reason why there would be a dependency conflict. @MCRE-BE checked with the mamba solver and that seemed fine, and concluded tentatively that there might be sth with the conda-forge solver that we cannot check: https://github.com/conda-forge/skbase-suite-feedstock/pull/51
possibly related: two weeks ago, sktime triggered a 3.12 rebuld here: https://github.com/conda-forge/sktime-feedstock/pull/123 - the 3.12 tests are failing, but not 3.8-3.11

dhirschfeld commented 5 months ago

See https://github.com/conda-forge/skbase-suite-feedstock/pull/49#issuecomment-2116384329

fkiraly commented 5 months ago

Ah, thanks! Did not see this because on a closed PR.

I think scikit-base could be noarch: python which would resolve this issue.

You'd need to put in a PR adding noarch: python to all build sections, removing all cross-python_ lines and all comment selectors (and then rerender).

Shall we try this?

I can make the PR, but I thought maybe you want the honour so the fix is in the record with your name, @dhirschfeld.

dhirschfeld commented 5 months ago

I won't have time to put in a PR, but might be able to review it.

fkiraly commented 5 months ago

Sure - first attempt: https://github.com/conda-forge/skbase-suite-feedstock/pull/52 FYI @dhirschfeld

MCRE-BE commented 5 months ago

Can be closed now no?

fkiraly commented 5 months ago

I would say, let's keep it open until the next release has gone through. Only then we will know that this has been solved without problems...

fkiraly commented 5 months ago

looks like all is fine. Thanks for your help, @dhirschfeld, @MCRE-BE!

Reference: https://github.com/conda-forge/skbase-suite-feedstock/pull/66

sktime / skbase

[MNT] `conda` release - compatibility with Python 3.12 #330