Closed seisman closed 4 years ago
Continuing on the discussion from #397, I'm just throwing out some options here. In order of least to most work, do we want to:
1) Just test against master on Linux (i.e. only edit the .travis.yml
file), basically #261 all over again.
2) Test on all platforms (Linux, MacOS, Windows and edit both .travis.yml
and .azure-pipelines.yml
3) Refactor our test suite completely, and maybe consolidate our tests into one platform in one go:
a) Follow core GMT
and put everything in the azure-pipelines.yml file. See https://github.com/GenericMappingTools/gmt/pull/988.
b) Go the Github Actions build matrix route, to test multiple platforms/Python versions. See also https://github.com/marketplace/actions/setup-miniconda.
As a side note, do we want to be testing for compatibility against multiple GMT versions in the long run (e.g. 6.0, 6.1, 6.2, etc?).
Ideally, we should test multiple GMT versions for compatibility. However, it seems difficult to maintain.
Sometimes, different GMT versions may produce slightly different images (e.g. a bug in 6.0.0, but fixed in 6.1.0), but the pygmt testing mechanism can only have one baseline image. Thus some pygmt tests must be failing for at least one GMT version.
Currently, we test pygmt with GMT 6.0.0rc4. I think we also need to test pygmt with the GMT master branch, to make sure changes in the core GMT don't break pygmt.