Julien Jerphanion
Julien Jerphanion
This issue is probably related to other ones on Windows, like https://github.com/mamba-org/mamba/issues/1106.
I would recommend considering performance improvement based on single threaded executions profiling records (one better optimise what one can measure). `py-spy`, `perf(1)`, `samply` with speedscope or Firefox profiler are handy...
It might also be worth understanding whether free-threading builds helps with some GIL-contention we observed in the past (see https://github.com/scikit-learn/scikit-learn/discussions/20651).
I confirm that I can reproduce with 2.4.0.
We cannot reproduce this issue with 2.0, please update to 2.0. Feel free to reopen an issue on `mamba-org/mamba` with a reproducer if you still have this issue with 2.0....
_A priori_, I do not think we are against the support of any 32bit architecture, it is just not the current priority. https://github.com/mamba-org/mamba/pull/3707 might interest you.
Thank you for this contribution, @henryiii. Someone with a machine on macOS 13 should dive into this problem. Note that we are only maintaining the distributions of mamba and micromamba...
> So far it looks pretty easily fixable, I think? Maybe just adding some explicit types instead of using the deduction guidelines. Yes, definitely. New GHA workflows can be set...
@henryiii: using macos-15 resolves the problem but it seems that homebrew's CI uses macos-13, macos-14 and macos-15. Must it use macos-13 and macos-14? I.e. would you accept those changes?
We don't have a macos-13 machine to debug, so we will wait.