gh-80010: Expand fromisoformat to include most of ISO-8601

[pganssle]

This should cover all of ISO-8601 except for fractional non-second components.

@godlygeek Would you mind taking a look?

Note: Currently the tests are mostly written as hypothesis tests using the stubs from #22863. Before merge we can try to refactor those out into a big matrix of examples, but for now it's useful to know that we have good coverage of the enormous state space here.

#80010

To Do:

• Add news entry
• Update docstrings
• What's new entry

[cpython-cla-bot]

All commit authors signed the Contributor License Agreement.

[pganssle]

I've added an explicit test matrix that doesn't rely on the hypothesis stubs. Feel free to ignore them as part of the review.

[JelleZijlstra]

A few comments, I'll have to study the C code more closely.

[orent]

Is there a way to limit this to only the rfc3339 subset? The full set of iso8601 features is not always appropriate.

[pganssle]

Is there a way to limit this to only the rfc3339 subset? The full set of iso8601 features is not always appropriate.

This sort of thing is the reason we delayed this expansion — the scope can easily grow if we do not constrain it to be something very specific, and we couldn't come up with a good way to allow users to express their preferences about what would be parsed. The approach we settled on is that fromisoformat would have a broad scope and not be configurable, and we will leave it to another library (or possibly another function) to allow further configuration.

I don't think it would be completely absurd to have a .fromrfc or fromrfc3339 method, though also RFC 3339 is so much more constrained than ISO 8601 that it's much easier to parse with something like a regex.

[jerub]

Thanks for asking me to review this change. I've made a few minor notes.

Great to see that we're building these batteries included. It would also be good to have a really solid RFC3339 parser as well. ISO 8601 has the mindshare because of the name, but RFC 3339 is the quality standard that people actually want.

[orent]

I don't think it would be completely absurd to have a .fromrfc or fromrfc3339 method, though also RFC 3339 is so much more constrained than ISO 8601 that it's much easier to parse with something like a regex.

Instead of adding yet another method - how about making the constructor accept a string? Many other builtin types have a constructor that accepts a single string argument and round-trips their __str__ (which may or may not be the same as their __repr__).

So my suggestion is for the constructor to accept the __str__ which is basically rfc3339 (possibly with a different separator) and the .fromisoformat() method would accept the larger set of ISO8601 values.

[pganssle]

Ok, I think we're down to just nits and other things that can easily be taken care of after feature freeze. I'm going to merge this.