Re: intl patches (18)

[Bruno Haible <haible at ilog dot fr>]

Ulrich Drepper writes:

>  patch 7: _tolower might be as efficient with glibc but not elsewhere.
>           If I'd care about those platforms I wouldn't have applied the
>           patch.

_tolower is nowhere documented. On those platforms, I don't want to
call an undocumented function even if it's more efficient.

>  The K&R changes:
> 
>           I've applied them only because this is still current practice
>           in most GNU utilities.  But we have to face C99 compilers hopefully
>           soon and then the K&R syntax is illegal.

In C99, K&R syntax is valid, but "obsolescent". See sections 6.9.1.6
and 6.11.7.

>           Saying "we have to do this because gcc must be compilable with
>           K&R compilers" is *not* a valid argument.  You can simply not
>           compile the NLS code if such a compiler is used.

It makes integration of gettext and gcc easier if there is no such
catch.

>  patch13: where is ICONV_CONST defined?

In config.h, generated by autoconf. It is not needed in _LIBC because
the _LIBC specific code uses __gconv, not iconv.

> patch 14, 15
>           If you want to maintain this (which you shouldn't
>           since you are only supporting enemies of free software)

For Win32, I agree with you, but not for the others. DOS is an "enemy
of free software" just as much as AIX or Solaris - which we attempt to
support. And emx+gcc for OS/2 is entirely GPLed.

>  patch14: too messy.  I'm not willing to have any such crap in glibc.  The
>           code is not intended to run on these degenerated platforms in the
>           first place. ...  create
>           a patch which adds to glibc's code only something like
> 
>           #ifndef SOME_MACRO
>           # define SOME_MACRO default value for reasonable systems
>           #endif
> 
>           and uses of these macros

This complicates the Makefile or autoconf stuff a lot more than a
simple #ifdef.

Bruno