On 09/20/2010 08:24 AM, Charles Wilson wrote:
On 9/20/2010 4:15 AM, Peter Rosin wrote:
But is it really a requirement? If you are referring to libtool commit
v2.2.10-92-g92d24b3, then that patch adds an /optional/ dependency on
autobuild. If you are a maintainer doing a release of libtool, it
probably is a requirement, but are you not simply building the released
version (perhaps with a few patches)? I know that I certainly have built
git libtool many times when working on it, and I do not have autobuild
installed.
cygport -- unless overridden -- always runs 'autoreconf' on a package.
autoreconf runs aclocal, which means autobuild.m4 must be available;
hence, *the way we routinely (re)build libtool*, autobuild is required.
No, if I understand Peter's comment correct, autobuild is NOT required
for libtool, even with an autoreconf. Why? Because the original
libtool patch at commit 92d24b3 made the use of AB_INIT optional - it
only calls the macro _if_ autobuild.m4 could be found in the first
place. When AB_INIT is not found, the only change is that the generated
./configure file no longer outputs magic strings that can be recognized
by an autobuild daemon; missing strings merely makes it harder to submit
build results to an autobuild daemon, but has no effect on how libtool
itself behaves. Reduced configure output is not a build dependency;
that would be the case only if the act of running autoreconf on libtool
flat-out fails if autobuild was not present.
Furthermore, I just sent a patch to libtool that avoids the need to have
autobuild installed in the first place:
http://thread.gmane.org/gmane.comp.gnu.libtool.general/11095/focus=10741
So, I still think that providing autobuild as a cygwin package makes
sense, but since I'm arguing that it is NOT a build requirement for
libtool, it means we need a previous distro link or the full 5 votes.
We're at 2 now; anyone else want to chime in?
--
Eric Blake ebl...@redhat.com +1-801-349-2682
Libvirt virtualization library http://libvirt.org