libc++ applies the two-digit logic even when one digit is read:

https://github.com/llvm/llvm-project/blob/b8d38e8b4fcab071c5c4cb698e154023d06de69e/libcxx/test/std/localization/locale.categories/category.time/locale.time.get/locale.time.get.members/get_year.pass.cpp#L40-L48

I'm not sure I agree with that, but should be go with it for compatibility?

libc++ applies the two-digit logic even when one digit is read:
I'm not sure I agree with that, but should be go with it for compatibility?

I agree that seems questionable. I'd recommend checking libstdc++'s behavior (not their source code) - if it does the same thing, then we may as well avoid implementation divergence. If libc++ and libstdc++ disagree, I think the compatibility argument is less compelling, and we should instead probably skip the affected test for now, and report it to libc++.

The libc++ logic appears to be as follows:

• %y: read up to 4 digits, apply 2-digit logic regardless of how many digits are read.
• %Y: read up to 4 digits, no post-processing
• time_get::do_get_year: same as %y

MSVC is:

• %y read up to 2 digits, apply 2-digit logic
• %Y: call get_year
• time_get::do_get_year: read up to 4 digits, apply 2-digit logic regardless of how many digits are read.

Here's the behavior of %y that I'm seeing (https://godbolt.org/z/P5zcjMb35):

There's unanimous agreement about applying the 2-digit rule for single-digit results, so I'll follow that.

More surprising to me is that both libc++ and libstdc++ will read more than two digits for %y. If I were working from a blank slate, I'd propose:

• %y read up to 2 digits, always apply 2-digit logic
• %Y: read up to 4 digits, no post-processing
• time_get::do_get_year: read up to 4 digits, apply 2-digit logic if two or fewer digits read (including leading zeros).

I think this is close to libstdc++ behavior, except for how many digits are read for %y. I don't see anything in the older or newer strptime wording that supports reading more than two digits.

We talked about this at the weekly maintainer meeting and @CaseyCarter, @strega-nil-ms, and I all agree with your "If I were working from a blank slate" proposed behavior. 👍

Looks great - I pushed changes for a few tiny things I noticed (followed by a trivial comment change to rerun the CLA Bot).

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

Thanks for investigating and fixing this runtime correctness bug while preserving ABI! 🐞 🔍 😻

I informed libstdc++ and libc++ developers: GCC-105450 and LLVM-55220

Oops, gcc already knew about it: GCC-103612 :)