Re: support cross multilibs

[Ralf Wildenhues]

Hello Geoffrey,

Apologies for the delay.

* Geoffrey Keating wrote on Fri, Sep 15, 2006 at 02:49:05AM CEST:
> Sometimes, a target has a multilib which doesn't actually work on
> host for that target.

> So, I'd like to put in this patch, which adds a little code to
> AM_MULTILIB to make non-default multilibs possibly be considered to be
> cross compilation.  autoconf will check whether compiling actually
> fails.

There are some things about this patch that I don't grok:
I was first going to complain that, since the first compiler test that
is invoked in a configure.ac (typically AC_PROG_CC) tests executability,
AM_ENABLE_MULTILIB would have to come before that, so that
$cross_compiling is properly validated there (an `AC_BEFORE([$0],
[AC_PROG_CC])' would have come in handy here), but when testing that
change on automake/tests/multlib.test, it leads to weird issues I
haven't fully analyzed yet.

Anyway, then I learn about AC_NO_EXECUTABLES resp. GCC_NO_EXECUTABLES
and see that GCC not only has much more magic in place to work around
things Autoconf does differently, also it seems in GCC's configure.ac
scripts AM_ENABLE_MULTILIB comes early enough, before compiler tests.

So to conclude: I guess the patch is probably not wrong.  And since GCC
is the most prominent user of multilib, the barrier to acceptance is low
on this matter.  OTOH, it clearly breaks the usage pattern shown in
automake/tests/multlib.test so I would prefer if either somebody more
authoritative could decide, or we can find a way to fix the test.

> I'd appreciate it if this could go on the 1.9.x branch as well as
> mainline.

I think Alexandre doesn't plan another 1.9.x release.

Cheers,
Ralf

> 2006-09-14  Geoffrey Keating  <address@hidden>
> 
>       * m4/multi.m4: Non-default multilibs may be cross compilation.