Install netcdf4 from conda-forge and pin netcdf4<1.7.3 in pyproject.toml#124
Install netcdf4 from conda-forge and pin netcdf4<1.7.3 in pyproject.toml#124valeriupredoi merged 5 commits intomainfrom
Conversation
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #124 +/- ##
=======================================
Coverage 74.05% 74.05%
=======================================
Files 11 11
Lines 2598 2598
Branches 406 406
=======================================
Hits 1924 1924
Misses 565 565
Partials 109 109 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
kmuehlbauer
left a comment
kmuehlbauer
left a comment
There was a problem hiding this comment.
Thanks @valeriupredoi 🤞
|
cheers @kmuehlbauer - still something 🐟 y going on with Python 3.10 though, am looking at it now |
|
A weird issue, cftime 1.6.5 was also released recently as well as h5py 3.15. Hope you find a working environment. I've given up over at h5netcdf and feed almost everything from conda-forge. |
|
the new wheel for 1.7.2 is causing it netcdf4-1.7.2-cp310-cp310-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl whereas the older wheel netCDF4-1.7.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl is working well 🤦♂️ |
yes! That's the way forward, we even have pip install --no-deps in a few other packages; I'll have to do the same here, but we'd probably have to move to pixi to support actual optional dependencies |
|
issues started coming in at netCDF4-python Unidata/netcdf4-python#1438 |
valeriupredoi
changed the title
|
@kmuehlbauer here's our current solution: install netcdf4 from conda-forge, and pin it in pyproject.toml (so eager pip doesn't grab 1.7.3, which is not yet on conda-forge, the 1.7.3 PR is hitting a broken build) - do you partake? 🍺 |
|
@valeriupredoi Maybe also add |
good call! Let me do it now |
done in 39b5de1 |
|
Great sleuthing @valeriupredoi, this looks like it's working properly. One thing I do not understand from the install log: Why do we end up with ancient s3fs and other packages versions. Is this really intended behaviour? |
lemme have a looksee, brb in 10min - good catch, I've not seen those yet |
|
boto3 and botocore are now at the latest version - s3fs is indeed ancient though
version mismatch with boto - we've been installing the superold s3fs for a long while now: it appears Amazon is not friends with fsspec much 😁 |
|
@kmuehlbauer fixed now, just don't ask me to run a |
kmuehlbauer
left a comment
kmuehlbauer
left a comment
There was a problem hiding this comment.
I'm in favour to get this in, thanks @valeriupredoi 🥇
|
cheers @kmuehlbauer - @davidhassell it's always a LOT better to get all these netCDFy/h5py-y deps off conda-forge - one HDF5 to link against to, and that's it; as for the pin in pyproject.toml that I put here, that is most definitely temporary, and we can remove it once the 1.7.3 conda-forge package is out (in a few days, it seems); as for the pin on s3fs, it appears there's a long standing version mismatch boto/s3fs, I'd rather have us a modern s3fs and have to pin for it TBF 🍻 |
2 participants
Description
Issues with the new released netCDF4==1.7.3 from PyPI
Closes #123
It is ONLY netcdf4==1.7.3 at fault since in an otherwise working env, simply installing 1.7.3 makes those tests fail:
Before you get started
Checklist