Well, that was the original code, but the real error was further down in the issue.

... and I didn’t understand well enough what was going on innpandas to write the test.

IMO there's nothing wrong with a more "integration test"-like thing that describes the desired behavior rather than a very specific test triggering the error.

When I wrote it, I checked it failed before the patch and passed after.

Sure but this test just errors because it’s not the desired behaviour and never worked. (You need to specify width with datetime units somehow). Well we could say that we should be able to infer the width from x which is what @anntzer latest or tries to do.

@jklymak I agree that a smoke test is not ideal, but it is better than not having a test.

I love smoke tests, its just the error was from #13186 (comment)... Hopefully your new test checks that use case...

This looks good. not sure whats going on w/ test cov....

What's confusing you about the coverage?

Ooops, it looks fine now