On Wed, Jul 12, 2023 at 11:49:59PM +0200, Bruno Haible wrote:
> Some more comments:
> * The encoding name "POSIX" is available as nl_langinfo (CODESET). Many
>   applications pass this encoding name to iconv(1) or iconv_open(3);
>   this is the main purpose of knowing the locale's encoding, after all.
> 
>   But if musl libc, or any other libc (Solaris or whatever), does the
>   same thing, but with a different mapping table than glibc has,
>   iconv implementations have a problem, because the name "POSIX" then
>   has different mapping tables on different systems.
Implementation survey via
	#include <langinfo.h>
	#include <stdio.h>
	int main(){
	puts(nl_langinfo (CODESET));
	}
and my original test program from https://bugs.debian.org/1017852 says:
  musl:             "ASCII"    (7f: 1, 007f     80: 1, df80	ff: 1, dfff)
  NetBSD 9.3:       "646"      (7f: 1, 007f     80: 1, 0080	ff: 1, 00ff)
  tribblix m25:     ‒ || ‒     (this is the "solaris or whatever")
  FreeBSD 13.0:     "US-ASCII" (7f: 1, 007f     80: 1, 0080	ff: 1, 00ff)
  OpenBSD 7.2:      ‒ || ‒
  MacOS:            ‒ || ‒     (uname says 19.6.0)
(notably, all of these are, strictly, wrong, but).

None of these implementations' iconvs have a POSIX/posix encoding, either;
I'm like mostly-sure this is just historical behaviour on the historical systems.

>   Suggestion: Change the encoding name from "POSIX" to "POSIX-GNU" or
>   something similar, to make it possible for iconv implementations to
>   support it in a future-proof way.
Given the above, I don't think there's much value to making the encoding
name weirder and longer, since the dust has long settled on this?

> * The identifier __gconv_btwoc_posix looks misspelled.
>   Suggestion: s/__gconv_btwoc_posix/__gconv_btowc_posix/
>   (The previous one, __gconv_btwoc_ascii, is misspelled as well.)
This explains why I didn't understand what it was trying to say lol;
applied, and propagated to __gconv_btwoc_ascii.

> * iconv/gconv_int.h: In the comment
>    identity-mapping bytes [0, 0x7F], and moving [0x80, 0xFF] into the end
>    of the Low Surrogate Area at [U+DC80, U+DCFF].  */
>   it's no longer the end of the Low Surrogate Area [U+DC00, U+DFFF].
>
>   Suggestion: Change the comment to
>    identity-mapping bytes [0, 0x7F], and mapping [0x80, 0xFF] into the
>    Low Surrogate Area at [U+DC80, U+DCFF].  */
Applied.

Will resend rebased for monday.