-
-
Notifications
You must be signed in to change notification settings - Fork 319
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
Change CI provisioning #4538
Change CI provisioning #4538
Conversation
Sphinx unpinned. lxml pinned as <5 rather than a specific version. Experimentally, add two workflows: Ubuntu 24.04 and Windows 2022 Signed-off-by: Mats Wichmann <[email protected]>
Also temporarily drops Ubuntu 22.04 from the matrix to get a quicker result. Use -j4 for build. Signed-off-by: Mats Wichmann <[email protected]>
Signed-off-by: Mats Wichmann <[email protected]>
The rest of the used action scripts are updated to today's current. The names of our actions are changed to match the task so it's easier to look at GH reports and see the difference (aka, don't call them all "run" like the examples). Signed-off-by: Mats Wichmann <[email protected]>
Signed-off-by: Mats Wichmann <[email protected]>
For the Windows workflow run, test In #4534, I removed the multiplier (i.e., effectively 1.0): # TODO: Reevaluate if having this part of the test makes sense any longer
# using precompiled headers should be faster
- limit = slow*0.90
+ limit = slow
if fast >= limit:
print("Using precompiled headers was not fast enough:")
print("slow.obj: %.3fs" % slow) In hindsight, it probably should have been changed to This test fails frequently for me when running tests in virtual machines. Even with the change, it still occasionally fails in one vm (i.e., pre-compiled headers slower). |
There are some mysterious fails due to adding the Windows runner, I guess I should have listed them here.
Many of these are mystifying. The The
The packaging tests fail with an error I've never seen,
... even though the packaging tool itself sets that. The pkg-config test fails because
There are a few others. Not able to spend time chasing these down at the moment. |
To address later. Signed-off-by: Mats Wichmann <[email protected]>
The pkg-config.py test contains
On windows, when using the default I could be wrong... |
Unlikely to be wrong. Yes, the tests are sometimes more ambitious than SCons is going to be with its limited PATH in the execution environment. I still don't know why perl gets involved. I've pushed a change to this PR which skips the 39 tests that failed last time, which has gotten us a "pass" this time. I'll record those in an issue and we can whack those off (and reduce the skip list) as motivation appears - no rush, really. |
The following could be wrong...
The
If the fully-qualified |
Never heard of Strawberry Perl, so I didn't realize it was a "thing", just thought the CI image had some weird paths. But it is a Thing indeed... http://www.strawberryperl.com/. So yeah, we found something in the more liberal test framework search, and then the necessary path wasn't in |
Problems found with thie Windows runner moved to #4539 |
Sphinx unpinned. lxml pinned as <5 rather than a specific version.
Experimentally, add two workflows: Ubuntu 24.04 and Windows 2022