address@hidden: Re: current glibc vs debian sid]

[Paul Eggert]

I'm forwarding this message containing a proposed change to Autoconf
to work around some problems that the glibc developers have run into.

The original of this message is in:
<http://sources.redhat.com/ml/libc-alpha/2002-10/msg00347.html>
The original glibc problem that started the thread is in:
<http://sources.redhat.com/ml/libc-alpha/2002-10/msg00303.html>

------- Start of forwarded message -------
From: Franz Sirl <address@hidden>
Date: Fri, 18 Oct 2002 22:02:30 +0200

On Friday 18 October 2002 20:25, Paul Eggert wrote:
> > Date: Fri, 18 Oct 2002 16:44:35 +0200
> > From: Franz Sirl <address@hidden>
> >
> > I've sent patches to the autoconf people a while ago, but it seems
> > they haven't been applied yet. A lot of autoconf stuff may be bitten
> > by this bug, not just glibc.
>
> Can you please give more detail about those patches?  I briefly
> searched the Autoconf archives for messages from you about ppc, but
> found only
>
> <http://mail.gnu.org/pipermail/autoconf-patches/2000-January/003661.html>
>
> and this doesn't seem to be relevant.

No, it was earlier in 2002 
http://mail.gnu.org/pipermail/autoconf-patches/2002-May/008371.html , but it 
was rejected back then, cause $build/$host/$target are deprecated. I even 
asked you about suggestions for a different solution.

Well, I just waded thru all this again and here a short summary.

Basically the semantics for

        ./configure some_alias

have changed heavily with autoconf-2.50 and later. Whereas in earlier autoconf 
the meaning was "build NATIVE with $build_alias/$host_alias/$target_alias == 
some_alias" it now means "build CROSS with 
$build_alias/$host_alias/$target_alias == some_alias". I think it would have 
been better to completely reject the old syntax...

Now, let's see how we are supposed to do this in autoconf-2.50+. The docs tell 
us to use --build, --host and --target, with --target only to be used with 
compilers (GCC canadian cross) and using --host=my-host-alias triggering a 
normal crosscompile (this is what changes the semantics like described above) 
to "my-host-alias".
Let's leave out the "canadian cross" thing for now, that gives us:

        NATIVE: ./configure --build=somebuild_alias
        CROSS: ./configure --build=somebuild_alias --host=somehost_alias

The problem here is that the NATIVE case doesn't set+replace $host_alias or 
$target_alias, but things like binutils and GCC rely on $target_alias being 
set+replaced even for NATIVE.

After thinking about it, I'm suggesting the following for autoconf:

 - reject the "./configure some_alias" syntax
 - handle $build_alias/$host_alias/$target_alias like $build/$host/$target, 
meaning that $build_alias defaults to config.guess, $host_alias defaults to 
$build_alias and $target_alias defaults to $host_alias. Note that the setting 
of $host_alias/$target_alias from $build_alias shouldn't trigger the setting 
of variables like $ac_tool_prefix or $program_prefix.

Comments?

Franz.
------- End of forwarded message -------