[RFC] setlocale(LC_ALL) vs setlocale(LC_NUMERIC)

[Martin Sebor]

Paolo Carlini wrote:
> Hi,
> 
> in order to fix (in 3_4-branch, at least), PR19642, I'd like to know 
> whether we
> can assume the below to succeed in "most" libc implementations: seems the
> case for glibc, but don't believe it's C standard mandated behavr:
> 
>  char* save = setlocale(LC_ALL, "de_DE");
>  assert( !save || !strcmp(setlocale(LC_NUMERIC, NULL), "de_DE"));
> 
> assuming, of course, setlocale(LC_NUMERIC, NULL) returns something
> like "de_DE", or "C", and so on, we can test at the outset for that.

I expected to find at least some locales where this wasn't true
(e.g., those that use aliases like "german" or "de" for "de_DE)
but I ended up with none on any of the machines I tried (which
included AIX, HP-UX, Linux, Solaris, and Tru64). So while I
wouldn't want to bet my paycheck on it I think it might be
a pretty safe assumption to make in a test suite.

Martin