autoconf for type sizes

[Zack Weinberg]

On Tue, Mar 13, 2001 at 12:02:52PM +0100, Akim Demaille wrote:
> >>>>> "Alexandre" == Alexandre Oliva <aoliva@redhat.com> writes:
> 
> Alexandre> In any case, I suppose it would be worth to borrow CVS
> Alexandre> autoconf's _AC_COMPUTE_INT framework, instead of
> Alexandre> re-inventing the wheel in GCC.
> 
> Is there any good reason not to use CVS Autoconf?  I mean, let's
> consider we release it now, or RSN.  So how about moving to 2.49d,
> which I will `release' today, so that we know if we need to change
> things for GCC in 2.50.
> 
> I fail to understand why people don't want to give a try to 2.49d.
> It's only delaying problems.  What difference does it make to find a
> problem with 2.49d as opposed to 2.50?

I'm going to ignore the lengthy flame war and just point out a few
practical issues unrelated to whether or not autoconf CVS is reliable.

We can't switch over to 2.50 for real, even on the mainline, until it
is released for real, because that'd force everyone working on the
mainline to track autoconf CVS snapshots.

I do understand and sympathize with your desire to have 2.50 tested by
gcc, which is quite demanding, before you release it.  Since I'm deep
in this stuff anyway, I'd be willing to chip in.

I noticed that 2.50 suggests the use of configure.ac instead of
configure.in for the input file.  GCC's configure.in ought to be
rewritten from scratch, and this looks like a decent excuse.

I'm nervous about backward compatibility.  Will arbitrarily ugly 2.13
configure scripts work with 2.50 without change?

zw