-- update --
hmm
#2619 (comment)

-- update --
Still, is it possible to replace bad_variant_access exception with something else, like

[[noreturn]] static void _Xbroken_iterator() {
      _Xinvalid_argument("the iterator has been broken due to an exception");
}

If this makes sense, we can move bad_variant_access back to <variant>.

(I think we can also add [[unlikely]] for these _Variantish_state::_Nothing branches.)

If this makes sense, we can move bad_variant_access back to .

Roughly like this: a9190c7; is this a valid approach?

I don't really care how a valueless_by_exception iterator should behave, but if we decide to not throw bad_variant_access (thus not equivalent to get and visit), we should probably ask LWG to make it standard.

Hmm, I wasn't aware the methods should behave as if operating on a real variant.

... So there is nothing changeable about <variant> in this pr.

@CaseyCarter I made a new push to replace _THROW(format_error(...)) with widely-applied _Throw_format_error(...) in <chrono>. (And I begin to feel uncertain about the relocation of _Throw_bad_array_new_length. It seems _Throw_something is often defined closely after the definition of something; though bad_array_new_length is special as it is conditionally defined there...)

(And I begin to feel uncertain about the relocation of _Throw_bad_array_new_length. It seems _Throw_something is often defined closely after the definition of something; though bad_array_new_length is special as it is conditionally defined there...)

I suppose it's a question of whether we consider _Throw_meow to be a local helper function that should be defined near its uses, or part of the API of meow which should be defined near the definition of meow. Maybe we should (in a separate change) go all-in on "part of the API" and make them static member functions meow::_Throw?

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

🧹 ⚠️ 😻