-
Notifications
You must be signed in to change notification settings - Fork 10
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
fix: Guard against removing package while uploading to it #64
fix: Guard against removing package while uploading to it #64
Conversation
ab1b478
to
f87a5b5
Compare
@kenodegard @dholth as you know the Anaconda Cloud web APIs better than I do, if you would have time to review the changes to |
@scientific-python/nightly-wheels-developers If it would be possible to get a review early this week it would be great to get this in so that I can create a new patch release with this. (I'm assuming when I tag @scientific-python/nightly-wheels-developers that everyone on the team gets a notification except me (the tagger), but I'm not sure if that's correct or not.) |
* On Anaconda Cloud as of 2024-01-04, if a wheel is being uploaded to a package, but the package only has one wheel in it and is of the same name as the uploaded wheel, Anaconda Cloud will overwrite the file by _removing_ the file from the package index. However, when this happens it removes the entire package, and then the wheel that is in the process of being uploaded has no destination and the upload fails. To guard against this, ensure for each package that has a wheel being uploaded if: - there is only one release for the package - and only 1 file for that release - and the upload target wheel has the same name as the file - that the file (and so the package) is removed in advance of the upload. * To make filtering names and versions from wheels easier, add a get_wheel_name_version function that uses as regex to lazily capture the package name as well as the version and then return these. - Examples of this working: * "matplotlib-3.9.0.dev0-pp39-pypy39_pp73-win_amd64.whl" matplotlib 3.9.0.dev0 * "scikit_learn-1.5.dev0-cp39-cp39-win_amd64.whl" scikit_learn 1.5.dev0 * "scipy-openblas64-0.3.26.186-py3-none-macosx_10_9_x86_64.whl" scipy-openblas64 0.3.26.186 * "awkward_cpp-29-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl" awkward_cpp 29 * As Anaconda Cloud normalizes package names, also distinguish between basename and package name to try to make things easier to keep track of when normalizing. * Example: - basename: test_package-0.0.1-py3-none-any.whl - basename prefix: test_package - package_name: test-package
f87a5b5
to
c4a9b40
Compare
@scientific-python/nightly-wheels-developers Ping again on this, as given that apparently merging now requires a minimum of 1 review this can't get merged until someone reviews. |
Thanks @jarrodmillman! I'm going to relock, merge, and then prep a new minor release as this add behavior without introducing breaking changes. |
* As curl and jq are now used in cmd.sh, they need to also be added to the conda environment.yml. * The lower bounds are chosen as the latest values, but are not motivated by known problems. * Rebuild the lock file.
* Add a test that triggers the conditions for removal of a package from Anaconda Cloud in advance of its upload to avoid an error.
c4a9b40
to
c37220f
Compare
Resolves #57
On Anaconda Cloud as of 2024-01-04, if a wheel is being uploaded to a package, but the package only has one wheel in it and is of the same name as the uploaded wheel, Anaconda Cloud will overwrite the file by removing the file from the package index. However, when this happens it removes the entire package, and then the wheel that is in the process of being uploaded has no destination and the upload fails. To guard against this, ensure for each package that has a wheel being uploaded if:
upload.
To make filtering names and versions from wheels easier, add a
get_wheel_name_version
function that uses as regex to lazily capture the package name as well as the version and then return these."matplotlib-3.9.0.dev0-pp39-pypy39_pp73-win_amd64.whl"
matplotlib
3.9.0.dev0
"scikit_learn-1.5.dev0-cp39-cp39-win_amd64.whl"
scikit_learn
1.5.dev0
"scipy-openblas64-0.3.26.186-py3-none-macosx_10_9_x86_64.whl"
scipy-openblas64
0.3.26.186
"awkward_cpp-29-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl"
awkward_cpp
29
"awkward-2.6.1-py3-none-any.whl"
awkward
2.6.1
As this requires
curl
andjq
also add these to the environment and relock.Add a extra upload in CI to thest this behavior.
In the future there will hopefully be a fix on Anaconda Cloud's side given anaconda/anaconda-client#702 but we need a solution now.