<syncstream>: is_nothrow_move_assignable_v<osyncstream> is squirrelly 🐿️

[StephanTLavavej]

WG21-N4928 [syncstream.osyncstream.overview] depicts basic_osyncstream& operator=(basic_osyncstream&&) noexcept;. However, it also depicts an exposition-only data member syncbuf_type sb; and [syncstream.syncbuf.assign] depicts basic_syncbuf& operator=(basic_syncbuf&& rhs); which is not noexcept (it calls emit()).

Our current implementation:

STL/stl/inc/syncstream

Line 348 in fbed7ea

basic_osyncstream& operator=(basic_osyncstream&&) noexcept = default;

STL/stl/inc/syncstream

Lines 371 to 372 in fbed7ea

	private:
	syncbuf_type _Sync_buf;

STL/stl/inc/syncstream

Lines 109 to 110 in fbed7ea

	basic_syncbuf& operator=(basic_syncbuf&& _Right) {
	emit();

It seems to me that there's an LWG issue here, and that operator=(basic_osyncstream&&) should not be marked noexcept in the Standard (and in our implementation).

Programmer-archaeology: LWG-3498 "Inconsistent noexcept-specifiers for basic_syncbuf" (already merged into the WP) removed noexcept from operator=(basic_syncbuf&&) because it calls emit(). It seems to me that operator=(basic_osyncstream&&) was simply missed at that time.

Noticed while reviewing #3314.

[frederick-vs-ja]

LWG-3867 has been reported.