Closed jaimergp closed 3 months ago
@jaimergp it looks like rc2 exposed other needed changes, see this Zulip message and subsequent ones (and napari/napari#6932
Thanks for the heads up @jni! I'll leave this on standby in the meantime.
Hm, now we have https://github.com/napari/packaging/issues/160, which suggests to pin numpy<2
for Python <3.10 (effectively, only Python 3.9?). I'm not sure if this is a problem that napari
itself has, or is more about being protective about some dependencies. Reading https://github.com/napari/napari/pull/6998, it looks like it has to do with scikit-image? 🤔 Why constrain here if the solver will handle that? I think it's ok to leave unpinned for all Python versions in condaworld, but lmk otherwise @jni
scikit-image does not have python 3.9 release compatible with numpy 2, but have python 3.9 releases without upper bound for numpy. That lead to pip installing numpy 2.0 and some old scikit-image.
Ah ok, I see, then we can repodata patch that in conda-forge, no probs!
Edit: I checked and in conda-forge there are no scikit-image releases that are compatible with py39 and do not pin numpy<2, so I think we are safe. The only ones with unpinned numpy are really old which are only available for Python <=3.6 anyway.