Our test suite runs on several versions of python. At present, we only see the test results for whichever run happens to fail first, as the others are cancelled.
This is problematic, as we may miss warnings and test failures that occur in some versions of python but not others. This PR configures github actions to run all jobs to completion even if there are failures.
In future we could re-enable "failfast" to save compute time, once the test suite is consistently passing on master. However, at present we should prioritise having good visibility of failures and warnings.
Side note: I think this was the root cause of a slight issue I had with PR #9 , which intended to fix some DeprecationWarnings. The test suite showed that the NumbaDeprecationWarnings had been fixed; however, the warnings still remained on higher versions of python, which I missed at the time. I'm making a second pass of a fix on #68.
Our test suite runs on several versions of python. At present, we only see the test results for whichever run happens to fail first, as the others are cancelled.
This is problematic, as we may miss warnings and test failures that occur in some versions of python but not others. This PR configures github actions to run all jobs to completion even if there are failures.
In future we could re-enable "failfast" to save compute time, once the test suite is consistently passing on master. However, at present we should prioritise having good visibility of failures and warnings.
Side note: I think this was the root cause of a slight issue I had with PR #9 , which intended to fix some DeprecationWarnings. The test suite showed that the NumbaDeprecationWarnings had been fixed; however, the warnings still remained on higher versions of python, which I missed at the time. I'm making a second pass of a fix on #68.