We have a regression test for DevCom-538510 (internal VSO-850856) "std::filesystem::weakly_canonical erroneously throws an exception":

This has regressed on the upcoming Win11 24H2, where we throw an exception (again). The relevant code is:

After calling _Canonical(), anything other than _Err == __std_win_error::_Success or __std_is_file_not_found(_Err) is considered a hard error (which will cause the non-error_code overload to throw a filesystem_error).

In current OSes, the fileWithSuffix scenario results in _Invalid_name here. (Locally verified on Win11 23H2. Presumably we get the same behavior in the GitHub PR/CI system running Server 2022 21H2, although I didn't verify that.)

In Win11 24H2, we get _Directory_name_is_invalid, which isn't recognized by __std_is_file_not_found. 💥

The fix is backwards-compatible and a simple one-liner: add this error code enumerator to the list of synonyms that __std_is_file_not_found recognizes. The plain meaning of _Directory_name_is_invalid is clearly a subset of _Invalid_name.

I've manually verified this (with significant but perhaps not surprising logistical difficulty) in an Azure VM with Win11 24H2. Regular test coverage will be provided by the MSVC-internal test harness (which is how we found this), and eventually by the GitHub test harness once Server 2025 24H2 is officially released.

Additionally, this uncaught exception from a non-error_code overload manifested itself as an abort() with no useful logs. To aid in future investigations, I'm including a simple change to the filesystem test. Now, we wrap the whole test in try ... catch and print the contents of any exception before failing.