In the :TODO: section, I think adding explicit section and paragraph
numbers is misguided. The stable names should suffice, and are,
y'know, stable. IDK how it is in core, but in library they're always
telling people to remove the numbers because they might change between
the review of the wording and when the editor applies the wording.

In "Normalization Discussion": Detection of un-normalized text is
fairly straight-forward, and GCC 10 already produces a warning.
Normalizing to NFC is not much more difficult.

Those two things are essentially identical in my and ICU's
implementations, FWIW.

Also in that section: Therefore, normalization of identifiers must
happen before object files are emitted. it is either the
responsibility of programmers to provide normalized identifiers, and
compiler to either warn or error on non-normalized ones, or for
compilers to normalize Unicode input. For non-Unicode input, the
transformation to normalized form is the simple table based lookup
used to translate now.

I don't see anywhere that you explicitly state that the paper proposes
to make non-NFC identifiers ill-formed, except in the wording. I'm a
bit concerned that there might be confusion.

Zach


On Wed, Apr 8, 2020 at 11:46 AM Steve Downey via SG16
<sg16_at_[hidden]> wrote:
>
> Draft R3 on isocpp.org for discussion.
>
> https://isocpp.org/files/papers/D1949R3.html
>
> There are some bits of wording that need fixing up that I plan to address before the 15th mailing, so having multiple P revisions seems overkill.
> --
> SG16 mailing list
> SG16_at_[hidden]
> https://lists.isocpp.org/mailman/listinfo.cgi/sg16