Em qui., 18 de mai. de 2023 às 16:56, Tom Honermann
<tom_at_[hidden]> escreveu:
> I don’t know what you mean by this. There are multiple implementation strategies, some of which have been in use for many years.

I mean importable headers as specified in C++20 is not implementable.
The things that exist all rely on an understanding that "source
inclusion" is still the "canonical" interpretation of the code, but
they perform the separate translation of the headers as an
optimization. If we unwind the specification to simply allow that
optimization, we don't really need to say anything else. But I suggest
we continue the discussion about that in the dedicated thread I
started (so folks don't miss it because it was not about the paper).

daniel