NLS code completely hosed

[Philipp Thomas]

* Richard Earnshaw (rearnsha@arm.com) [20000603 16:15]:
 
> I can't believe the NLS code has been left enabled for 4 days now, given 
> just how broken it seems to be.  

Well, you could always configure with --disable-nls, to switch it off. Sad
as the outcome is at the moment, this is exactly one of the reasons why I
pushed for enabling it by default: to give the stuff broader exposure and
kill the bugs present in that code.

> 1) As has been mentioned before, localealias.c blithely defined memmove in 
> terms of bcopy, but then assumes it can use the result.

> 2) It tries to include config.h from the gcc directory, but that file is a 
> no-op if IN_GCC is not defined.
> 
> 3) It isn't passed the list of include directories, so even if IN_GCC is 
> defined, it can't read the various include files that it does need.
> 
> 4) Even when I do fix the command in gcc's Makefile, to pass the correct 
> subdir flags, intl/Makefile, tries to compile localealias.c twice and for 
> one of the invocations it is still using the wrong flags (I think is a 
> parallel make bug in the makefile)
> 
> The patch below sort of makes sure we pass the correct flags to the 
> sub-make, but even when we do, we still get link errors when trying to 
> build cc1chill.  I've no idea what is going on now...

I have pending patches to solve these problems. I just didn't have the time
to thoroughly test all the stuff. Amongst other, my patches do is:

 - introduce IN_INTL, which will only make auto-host.h be included, as
   that's all the libintl stuff needs

 - fix localealias.c to not use the result of memcpy.
 
I expect to check these in either today or tomorrow.

Philipp

-- 
Philipp Thomas <pthomas@suse.de>
Development, SuSE GmbH, Schanzaecker Str. 10, D-90443 Nuremberg, Germany

#define NINODE  50              /* number of in core inodes */
#define NPROC   30              /* max number of processes */
 	-- Version 7 UNIX fuer PDP 11, /usr/include/sys/param.h