CI: Running tests with xvfb for the clipboard to work #26932
Conversation
datapythonista
added
Testing
|
No idea why And it also worries me, that in my fork pipelines (I set it up to test changes to the CI), things fail that don't fail here. I feel like I was dealing with windows 95... :\ |
|
In the build |
|
I see that |
| @@ -46,6 +46,11 @@ do | |||
| fi | |||
|
|
|||
| PYTEST_CMD="pytest -m \"$TYPE_PATTERN$PATTERN\" -n $NUM_JOBS -s --strict --durations=10 --junitxml=test-data-$TYPE.xml $TEST_ARGS $COVERAGE pandas" | |||
| if [[ "$(uname)" == "Linux" ]]; then | |||
| # This is required for the clipboard to work (X virtual frambuffer implements the X server API, required by the clipboard tools, without running an actual X server) | |||
| DISPLAY=:99.0 | |||
There was a problem hiding this comment.
so u prob need to start xvfb much earlier and background it
There was a problem hiding this comment.
I'm using xvfb-run, which starts it for the command, and stops it when the command finishes. It's what xvfb's got in the azure servers. I guess something else can be installed, but I'll see what I can do.
May be time to try again to call the tests just once (not twice for single and multi). :)
There was a problem hiding this comment.
I think I’ve seen similar things for matplotlib headless tests. IIRC they add a 1-second sleep after this for the thing to start
|
Maybe try in a step before? On Travis, they explicitly say in their docs it needs to be run in the "before_script" section (so the section where the tests are run) |
… xvfb is still running, as it looks like
datapythonista
changed the title
|
I think the @jorisvandenbossche not an expert, but what I'm doing now is a bit different than what we've got in travis. In travis we just start the xvfb, and leaving it running. Here I use |
|
Confirmed that the problem is that I can do a sleep for now if you want to get this in faster, but I think the right solution is the approach in #24106 I see that in travis the same solution doesn't seem to work, and clipboard tests are skipped. :\ |
Codecov Report
@@ Coverage Diff @@
## master #26932 +/- ##
==========================================
- Coverage 91.87% 91.86% -0.01%
==========================================
Files 180 180
Lines 50716 50743 +27
==========================================
+ Hits 46593 46617 +24
- Misses 4123 4126 +3
Continue to review full report at Codecov.
|
Codecov Report
@@ Coverage Diff @@
## master #26932 +/- ##
==========================================
- Coverage 91.87% 91.86% -0.01%
==========================================
Files 180 180
Lines 50743 50743
==========================================
- Hits 46620 46616 -4
- Misses 4123 4127 +4
Continue to review full report at Codecov.
|
In the previous PR where I fixed those tests, I moved it back from run_tests.sh to before_script exactly because it was not working there (as said, it's in the documentation of travis it needs to be there). You're free to try to find a solution that works for both, but personally I wouldn't spend too much time on that if you have something that works for Azure. |
datapythonista
changed the title
|
Going for the simple hacking solution for now, so the tests start running. If I have time I'll give another try to call pytest just once, that will solve the problem for azure, and simplify the CI significantly. |
datapythonista
changed the title
… wait for previous run to finish
|
Closing this, as I think it make more sense to implement those changes in a fresh PR after #26949 is merged (which will simplify this significantly). |
git diff upstream/master -u -- "*.py" | flake8 --diffWith some luck this should make the clipboard tests work in azure. @jorisvandenbossche @jreback can you have a look once the CI finishes?