CI: revise the 'cygwin' job of the "Wrap releases" workflow #5289
Conversation
fingolfin
added
release notes: not needed
|
Excellent, the failure of #5285 is perfectly reproduced. |
|
So I've added a HACK commit which works around the issue currently troubling the package distro (i.e. it removes the broken symlink in the AGT package before creating the So in principle this seems ready to me, though I'd appreciate if it got some review; and of course that HACK commit should be removed, and the package distro be fixed instead (ideally by getting an update for the AGT package in). |
| # | ||
| # Furthermore, this would remove the need for the | ||
| # download_release_archive.py script, and if we would also remove the need | ||
| # for the upload_file_to_github_release.py script, as described below, |
There was a problem hiding this comment.
argh, ok, so we still need to make sure upload_files_to_github_release.py is available, or replace it (which I'd prefer anyway).
But also: this mistake of mine was not revealed because this code path is only tested for a release. Bad!
Perhaps we can always perform the upload -- but to where? Not sure this is possible. But perhaps we can do it at least for the scheduled daily jobs: we could let that (re)create a release nightly (i.e. delete it if it exists, then tag the commit from which the scheduled workflow is run, and upload there). Then we still notice failures relatively quickly
There was a problem hiding this comment.
Maybe also should get rid of make_github_release.py... I'll give it a spin
There was a problem hiding this comment.
Done, and verified on my personal fork of the gap repository (i.e. I made a dummy tag there and it successfully built and uploaded the .exe installer and the .sha256 file for it.
fingolfin
force-pushed
the
mh/refactor-release-ci
branch
from
0143357
to
322f8ca
Compare
First, we change the 'unix' job to always upload the .tar.gz it produces as an artifact. This puts a little extra burden on GitHub; but it lets us modify the 'cygwin' job to always use this artifact. This removes a major point in which our simulated releases differed from actual releases. This subtle difference already caused trouble for two releases. With this change, I hope it won't cause trouble a third time. As an added benefit, it also allows us to remove a bunch of custom code from the 'cygwin' job, and let's us avoid cloning GAP.
fingolfin
force-pushed
the
mh/refactor-release-ci
branch
from
322f8ca
to
6b9110d
Compare
fingolfin
force-pushed
the
mh/refactor-release-ci
branch
2 times, most recently
from
09ef8c4
to
60ecb8f
Compare
fingolfin
force-pushed
the
mh/refactor-release-ci
branch
from
60ecb8f
to
ce27182
Compare
|
Is there even aa cygwin version for ARM Windows? Anyway, if GitHub adds runners for ARM Windows, we can of course add that. The real reason for the matrix of course was that building these installers is soooo slooooooow. I've opened an issue about that |
First, we change the 'unix' job to always upload the .tar.gz it produces as an artifact. This puts a little extra burden on GitHub; but it lets us modify the 'cygwin' job to always use this artifact.
This removes a major point in which our simulated releases differed from actual releases. This subtle difference already caused trouble for two releases. With this change, I hope it won't cause trouble a third time.
As an added benefit, it also allows us to remove a bunch of custom code from the 'cygwin' job, and let's us avoid cloning GAP.
Motivated by issue #5285
Resolves #5011