This has not been apparent because the tests for this code are not testing what they think they're testing.

@gphelmsley Can you clarify what you mean by this?

Do you have a minimal reproducing example that shows what's happening and what you are expecting?

I've created a PR that fixes the issue, which I discovered while evaluating the test coverage for _strptime.

Certain scenarios of error messages were never being hit because the cascade was out of order, and the tests were not showing that because they were throwing a different ValueError than the one they were expecting to throw.

Hm, I was a bit confused by your wording here, because I am able to trigger all the errors just fine even before this PR, but I do think that even though this isn't necessarily fixing inaccurate error messages (all the error messages *are* accurate), you're right that the text of the messages does seem to indicate that the original authors intended a cascade ordering more like the one you've proposed.

I'll give this a more thorough review a bit later, thanks for working on this!

Ah yes, to be clear, I wasn't trying to suggest that the error messages themselves were wrong—just that they weren't triggering when the tests were expecting them to.

Some of the existing tests currently trigger the "unconverted data remains" ValueError from earlier in the method, but because the messages are not checked, that is not immediately obvious. I've also added new tests for additional scenarios that would presumably also be considered invalid, based on the existing ones.