LWG-3571 flush_emit should set badbit if the emit call fails

[StephanTLavavej]

LWG-3571 flush_emit should set badbit if the emit call fails

This was voted into the Working Paper in October 2021, see #2236.

[MattStephanson]

I was going to post this on the PR, but it's more about the issue itself, so I'll put it here.

The LWG resolution makes this an "unformatted output function", so it seems to need a sentry object and everything that goes with it from [ostream.unformatted]/1.

On the other hand, LWG-3570 makes basic_osyncstream::emit itself an unformatted output function, so constructing a sentry both here and in the emit call seems redundant. Both issues use the typical "After constructing a sentry object..." wording, which I read as requiring two separate sentry objects, unless there's an argument that the as-if rule applies.

[MattStephanson]

FYI, @fsb4000.

[fsb4000]

Thank you @MattStephanson ! I completely missed this sentence when doing the implementation.

But I think both sentry objects are needed because one sentry object is created in basic_osyncstream::emit but flush_emit calls _Basic_syncbuf_impl::emit

[MattStephanson]

But I think both sentry objects are needed because one sentry object is created in basic_osyncstream::emit but flush_emit calls _Basic_syncbuf_impl::emit

@fsb4000 I think you're right, I was confusing basic_osyncstream::emit (which is an unformatted output function) with basic_syncbuf::emit (which isn't). flush_emit calls the latter, so there's no issue of duplicate sentry objects.