how to patch configure.ac and not require autotools

[Dan Horák]

Ralf Corsepius píše v Po 12. 06. 2006 v 03:26 +0200:
> > them).  I'd suggest simply patching things and
> > BuildRequires: automake autoconf
> > and be done with it.
> 
> . and wait for blindly generated diff to break your package. 
> 
> 
> Pedantically speaking, the opposite is true. Modern automake-based
> packages support and ship "missing":
> c.f. /usr/share/automake-1.9/missing --help
> 
> I.e. the packager provides a *complete and functional* patch which
> breaks timestamps on generated files, and does NOT to have the autotools
> installed, the autotools will "touch the generated files", themselves
> 
> I.e. Actually, BuildConflicts: <autotools>
> and using missing would solve the problem.
> 
> 
> Less pedantically, if the default set of packages inside of the buildsys
> doesn't contain the autotools, and if packagers were required to provide
> complete patches, this issue becomes a non-issue. 
> 
> The down-side would be users rebuilding a package and having the
> autotools installed, would see the autotools be run, i.e.
> non-deterministic "user-builts".
> 
> Build qgit with the patch below and the autotools uninstalled to see the
> effects I am talking about.

I have tried the same patch when playing with the ways to solve this
problem and know what it is doing and this is also mentioned as an
option in my email that started this discussion.

When I summarize it
1. use BuildRequires: <autotools>
- generally can break the build due incompatibilities between author's
autotools and the build system's autotools

2. use BuildConflicts: <autotools>
- secure for the build system, but when the autotools will be in the
default set, then the build will fail (the buildsystem cannot uninstall
a package or can it?), dtto for users

3. use neither BuildRequires or BuildConflicts
- in the build system without autotools it will use "missing" and the
user is responsible for what is he doing

In this case I am sure, that patching both configure.ac and configure
doesn't require any changes to be propagated further, so I should use
point 3 and let "missing" do its work, if I understand it correctly.


		Dan