Calling autoconf in a spec.

[Ralf Corsepius]

On 07/04/2011 12:39 PM, Richard W.M. Jones wrote:
> On Sun, Jul 03, 2011 at 11:45:09AM -0400, Tom Lane wrote:
>> FWIW, I used to run autoconf during the builds of both the postgresql
>> and mysql packages for many years.  That eventually became unnecessary
>> in both cases, but I don't recall that autoconf changes ever caused me
>> any trouble.  (gcc, on the other hand, ...)
>
> We've been bitten with very old versions of autoconf (hello, RHEL 5).
> In those cases we've had to patch the generated files instead of the
> autoconf input files.  It's a pain in the derriere doing this.
Depends. In some cases it's difficult, in some cases it's simple
(Often it's just commenting out a few lines in both configure.ac and 
configure.)

> However this is not a reason not to rerun autoconf.  If rerunning
> autoconf is going to fail, then it will fail early and obviously.
Right, it often fails early and obvious, but there are cases "brute 
force" re-generated scripts produce bogus (broken) results.

Cases from the latter group often are very to detect often get away 
unnoticed for long (Until somebody notices them ...) - These are the 
critical ones.

Some historical examples:
https://bugzilla.redhat.com/show_bug.cgi?id=449944
https://bugzilla.redhat.com/show_bug.cgi?id=568039

In both cases, people running autoreconf with these versions would have 
silently produce miscompiled and possibly dysfunctional packages. 
Fortunately, both issues only tangentially hit Fedora (BZ449944 didn't 
affect Fedora at all, use cases which would be hit by BZ#568039 are rare).

> This problem isn't likely to affect Fedora because Fedora usually has
> the latest autotools.
It affects Fedora packages with dead upstreams or packages which prefer 
to stick with outdated tools and which rely on certain outdated auto* 
constructs/feature (e.g. gcc, binutils, .. firefox ...)

> Packagers should use their discretion (as in many other cases).
Full ACK.

Ralf