Re: contrib/gcc_update in non-english locales

[Paolo Bonzini <bonzini at gnu dot org>]

On 04/22/2010 09:21 AM, Dave Korn wrote:
> On 22/04/2010 06:17, Basile Starynkevitch wrote:
> > > > On 04/21/2010 06:25 PM, Basile Starynkevitch wrote:
>
> > > > > The contrib/gcc_update script does not seem to work in non-english
> > > > > locales (e.g. in a French UTF8 locale on Linux).
>
>    You never told us what the problem actually is

But I know. :-)  See the links that I posted when I read the reverse 
patch from Basile.

There are two issues:

1) collation.  [A-Z] and [a-z] match both uppercase and lowercase 
(except respectively "a" and "Z") because the collection order is 
aAbBcCdD...zZ.  In some locales they might even match a group of >1 
letters, e.g. ch or ll in Spanish.  This is the most common issue.

2) encoding.  In UTF-8, . would match only valid UTF-8 characters and 
again could match more than one byte.  Furthermore, the meaning of [à] 
is different in ASCII and UTF-8: again, UTF-8 would match more than one 
byte in this case and only in the specific order 0xc3-0xa0, while ASCII 
would match "half" of the UTF-8-encoded character only, and would match 
an unpaired 0xa0 too (invalid UTF-8, but e.g. valid Latin-1).  This is 
rarely a problem.

> > +LANG=C
> > +LC_ALL=C
>
> ... that if it's the encoding, that patch won't solve the problem on systems
> where the default encoding for the C locale is UTF-8 anyway; it would need to
> be something like this:
>
> > +LANG=C.ASCII
> > +LC_ALL=C.ASCII

No, C.ASCII is only supported by Cygwin as far as I know.  It works 
under glibc only by virtue of being an unknown locale (so that you end 
up using the default, which is C), but it's not portable.

C.UTF-8 in particular does not work under glibc (and making it the 
default in Cygwin was a veeeeery bad idea).

Paolo