Closed thermokarst closed 2 years ago
Hmm, these tests pass for me locally. I'll take a closer look at this tomorrow.
Is it possible this is a stochastic failure? Are the usage examples sorted at all?
You got it, @thermokarst. networkx doesn't guarantee deterministic results from the topological sort that orders these import statements, and I suspect that's what's causing these inconsistent test failures. For posterity, I'm going to split the exp
expression and check the two import blocks independently with no expectation of ordering.
There may be other tests impacted by this, but I'm inclined to knock them out as they present as problematic, rather than putting time into a non-deterministic which hunt™️.
I don't seem to have edit access to this PR, but if you cherry-pick the top commit from here (fceff7761e515f361051f1527c528d074ec7c469) it should drop the ordering in the test. I can push that up myself if you'd prefer not to deal with it.
I noticed the github actions workflow wasn't running on pushes to main so the first commit here updates that, and the second commit tweaks a test that is currently failing (probably bc ci didn't run on the related commits).