search

hpi-swa / smalltalkCI

Framework for testing Smalltalk projects with GitHub Actions, GitLab CI, Travis CI, AppVeyor, and others.
MIT License
96 stars 68 forks source link

Pharo tests fail on Windows #529

Closed fniephaus closed 3 years ago

fniephaus commented 3 years ago

smalltalkCI currently ignores failures for Pharo (and Moose) on Windows. If it doesn't ignore them, we get a couple of test failures:

image

If someone with access to a Windows machine could have look, that'd be great. There may be an issue with UUIDs?

theseion commented 3 years ago

Yes, the problem are UUID's. The UUID implementation is clock based, so the first 8 bytes are taken from the system clock. If the test runs fast enough, that value will not have changed yet.

There are a couple of different ways to fix this, e.g. "fixing" the test by introducing a delay, using more than 8 bytes for the Travis ID, copy the 4 bytes from the end of the UUID, which is random... What would you like? Ideally, we would just use full UUID's, IMO.

fniephaus commented 3 years ago

Copying from the end of the UUID sounds good for now. I don't like "random" delays in tests. And I remember that these Travis IDs are supposed to be short, otherwise the UI bails. Could you propose a PR?

theseion commented 3 years ago

Sure. Might not be today though.

fniephaus commented 3 years ago

There's no rush! Wouldn't have time to review and merge it today anyway ;)

fniephaus commented 3 years ago

Test failures are no longer ignored in Windows: https://github.com/hpi-swa/smalltalkCI/commit/46e8d5fb8e9239760f9e180c3c2aad507b38be29.