-
Notifications
You must be signed in to change notification settings - Fork 25
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Skip tests that require kerchunk #259
Conversation
@@ -0,0 +1,60 @@ | |||
name: min-deps |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There was probably a better way to do this than just making an entire extra workflow file but my github-actions-fu is weak
"xarray>=2024.06.0", | ||
"kerchunk>=0.2.5", | ||
"h5netcdf", | ||
"numpy>=2.0.0", | ||
"ujson", | ||
"packaging", | ||
"universal-pathlib", | ||
"numcodecs", |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The minimum set of dependencies should look like this, and you could probably even make universal-pathlib
optional for reading from non-remote files...
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Actually I re-added ujson
, because we need it for parsing existing kerchunk-formatted references.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Let's also leave the question of whether we really need fsspec/universal_pathlib for later, and just concentrate on removing kerchunk.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It might be nice to have more splits in the optional deps depending on the usecase, kind of like how Xarray handles it. We could have splits for different readers, writers and local/cloud usecases?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I was thinking the same, but kept this PR targeted for now.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks great @TomNicholas!
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Probably for another discussion, but could we have all deps and optional deps in pyproject.toml then the optional h5py
in a conda env?
Or maybe it's a good reason to transition to Pixi 🥇
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I know nothing about pixi, but would be open to that.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@maxrjones or @cisaacstern might have some real experience with it, but one really cool seeming thing is that you can have both conda and pip deps all in a pyproject.toml
"xarray>=2024.06.0", | ||
"kerchunk>=0.2.5", | ||
"h5netcdf", | ||
"numpy>=2.0.0", | ||
"ujson", | ||
"packaging", | ||
"universal-pathlib", | ||
"numcodecs", |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It might be nice to have more splits in the optional deps depending on the usecase, kind of like how Xarray handles it. We could have splits for different readers, writers and local/cloud usecases?
@@ -13,9 +12,17 @@ | |||
from virtualizarr.backend import FileType | |||
from virtualizarr.manifests import ManifestArray | |||
from virtualizarr.readers.kerchunk import _automatically_determine_filetype | |||
from virtualizarr.tests import has_astropy, has_tifffile, network, requires_s3fs | |||
from virtualizarr.tests import ( |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This is a great addition! Thanks for implementing it.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Another good idea just stolen from xarray 😅
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The source of all good code
All I've done here is skip all of those that currently require kerchunk, and add a new CI job to check that the others pass without kerchunk installed. I'm hoping that we can merge this PR then incrementally re-write all of the tests (apart from those in
test_readers/test_kerchunk.py
&test_writers/test_kerchunk.py
) to run and test the same functionality overall without requiring kerchunk.