Re: [E-devel] Improving the build system

[Kevin B. <co...@co...>]

Richard Torkar wrote:
> 
> 
> On Fri, 2005-02-25 at 14:36 -0500, Michael Jennings wrote:
> > On Friday, 25 February 2005, at 18:05:42 (+0100),
> > Richard Torkar wrote:
> *SNIP*
> > > so I'm just curious as to why this was a problem in this project. As
> > > always, I'm asking nicely. I'm not being arrogant - just curious and
> > > maybe stupid :-)
> >
> > 1.  Automatically-generated spec files often fail to end up in the
> > release tarball.  This prevents "rpmbuild -ta" from working.  Your
> > patch has exactly this problem.
> 
> I'm well aware of that fact. It was not the intention to make the
> spec.in full-fledged to begin with. If that is the purpose then I'd be
> more than happy to fix that.
> 
> > 2.  Automatically-generated spec files lull software maintainers into
> > a false sense of security and are generally left to stagnate.
> 
> Yes, unfortunately I've seen this as well...
> 
> > 3.  Spec files and the rpm dependency engine are not equivalent to
> > autoFUCK tools and cannot be treated as such.
> 
> Wholeheartedly agree.
> 
> > 4.  Making spec file changes, like making configure.in changes, can
> > have notable and widespread results and should thus be done as a
> > conscious choice.
> 
> See point 3, since it's the same.
> 
> > 5.  AutoFUCK tools generate files which differ between systems and
> > between builds.  If you have to change your spec file whenever you
> > build, you have issues.
> 
> See point 3, since it's the same.
> 
> What I wan't to do, and what I believe more people want, is to make spec
> creation a first class citizen in E's build system. Now if the dev team
> decides not to, then I have no problems with that. I'll just use
> Didier's rpms which he creates with some pearl magic etc. :)

Well, as the other guy who does .spec update once in a while, I'll vote
with Michael on this one.  Except for version automation, I don't see
much advantage and I agree with the objections.

-- 
Kevin