Re: Proposed Autoconf 2.50 path
On Sun, 27 May 2001, Sam Couter wrote:
> Marcus Brinkmann <Marcus.Brinkmann@ruhr-uni-bochum.de> wrote:
> > The lesson to learn here is that packages should not require libtool,
> > autoconf, or whatever at build time. Those tools are designed to be run at
> > source preperation time, so it is the responsibility of the package
> > maintainer to run them before packaging the source.
> So... should my packages not require gcc or ld to build either?
> Sorry, I don't agree with your argument. A generated file is a generated
> file, and if the upstream source doesn't have it (or if you've patched the
> files which are input source for the generated file), then you need to
> generate it.
But the configure script is an *architecture-independent*, generated file.
Most other files that are generated at build-time are both
architecture-dependent and dependent on the specifics of the environment in
which they're being built. I don't see any reason that developers *must* not
include the output of autotools in CVS. Developers always have the option of
deciding amongst themselves to mandate a particular version of autotools for
their project, and worst case, developers (at least the ones I've worked with)
are bright enough to recognize conflicts in the configure script when they
happen, and resolve them before committing to CVS.
Also, since many projects now have publically-accessible CVS repositories,
including a working configure script is going to save developers from fielding
a lot of questions from those who have arrived more recently at the cutting
edge.
And of course, since when the software is released (as a tarball/deb/etc) you
want there to be a working configure script within, it's good to be able to
test said script before you release -- which means making the exact version of
the script, as generated on the release manager's machine, available for
testing.
Steve Langasek
postmodern programmer
Reply to: