Re: [PATCH v7] POSIX locale covers every byte [BZ# 29511]

[Florian Weimer <fweimer@redhat.com>]

* наб:

>> Not sure what is more important here, musl compatibility or Python
>> compatibility.  Cc:ing Victor in case he as comments.  I should probably
>> ask on the musl list as well as how this divergence came to pass.

> I went for musl because (a) it's a libc not some random programming
> language, (b) putting the end of our domain at the end of the
> surrogates is more aesthetically and ideologically pleasing, and (c)
> there's marginal value of having both musl and glibc produce the same
> characters if you like save them as integers for some reason.
> But the choice of any range therein is pretty much editorial, I think.

Let's wait and see what the musl folks say.

>> This change definitely needs a NEWS entry.
> Something like this?
>   Deprecated and removed features, and other changes affecting compatibility:
>   * The default/"POSIX"/"C" locale's character set is now "POSIX",
>     instead of "ANSI_X3.4-1968"  this is a new fully-reversible
>     8-bit transparent encoding for compatibility with Issue 7 TC 2,

“POSIX Issue 7 TC 2”

>     identity-mapping bytes in the ASCII [0, 0x7F] range,
>     and mapping [0x80, 0xFF] bytes to [<U+DF80>, <U+DFFF>].

It should go into the major new features section, I think.

I would also say that POSIX no longer allows using UTF-8 for the C/POSIX
locale because the obvious question will be “why this custom encoding
and not UTF-8?”.  This new POSIX requirement is still a major
disappointment to me.

No need to repost for now.

>> > diff --git a/stdio-common/tst-printf-bz25691.c b/stdio-common/tst-printf-bz25691.c
>> > index 44844e71c3..e66242b58f 100644
>> > --- a/stdio-common/tst-printf-bz25691.c
>> > +++ b/stdio-common/tst-printf-bz25691.c
>> > @@ -30,6 +30,8 @@
>> >  static int
>> >  do_test (void)
>> >  {
>> > +  setlocale(LC_CTYPE, "C.UTF-8");
>> > +
>> >    mtrace ();
>> >  
>> >    /* For 's' conversion specifier with 'l' modifier the array must be
>> 
>> What's the rationale for this change?  If it is really required, you
>> must also update stdio-common/Makefile with a new dependency on
>> $(gen-locales).
> The test depends on the locale having a hole at 0xFF, cf. ll. 93-100:
>     /* Same test, but with an invalid multibyte sequence.  */
>     mbs[mbssize - 2] = 0xff;
>
>     ret = swprintf (result, resultsize, L"%.65537s", mbs);
>     TEST_COMPARE (ret, -1);
>
>     ret = swprintf (result, resultsize, L"%1$.65537s", mbs);
>     TEST_COMPARE (ret, -1);
> And this is the simplest way to ensure that, I think.
>
> Dependency added.

Right, makes sense.

Thanks,
Florian