Re: Building SVN source failures: VPATH builds; Cygwin

[Eli Zaretskii]

> Cc: address@hidden, address@hidden
> From: Hans-Bernhard Bröker <address@hidden>
> Date: Sat, 14 Jan 2017 16:31:51 +0100
> 
> >> I usually prefer sticking with the "no generated files in version
> >> control (except for rather exotic generators)" approach.
> >
> > That would require everyone who wants to build out of the repository
> > to have all the necessary tools of the correct versions installed,
> > which might not be a very user-friendly approach.
> 
> Git sources already routinely require people to have quite a collection 
> of tools --- more than distributed tarballs, anyway.  Walking along the 
> bleeding edge does require stronger shoes.

But the shoes don't need to be stronger than necessary, and requesting
them to be more than that is less user-friendly than possible.

> > I actually don't understand why this urge to remove generated files
> > from the repository, since VPATH should DTRT with stale files anyway,
> 
> The reason I started this thread was that the ones in place really 
> didn't.  The rules to generate fresh files weren't even run, because the 
> VPATH rules believed there was no need.  But the rules using them didn't 
> find them.  Worse yet, the gnulib configuration ruls failed _silently_, 
> causing rather weird compiler errors.

That just means the Makefile rules need some tweaking to DTRT.  The
solution is not necessarily removal of the files from the repository.

> I can see the point in keeping files in the repository that would 
> require somewhat exotic tools to update / regenerate.  I honestly can't 
> see automake or autoconf as exotic, though.

That you don't see them as exotic doesn't mean everyone else think the
same, or have them installed everywhere.  In particular, having that
on Windows is not a trivial thing.  (Yes, I do have them, but many
people cam up complaining that they couldn't set up these tools
correctly.)

> Building texindex.awk from ti.twjr uses jrtangle, which requires not 
> just any awk, but specifically GNU awk.  Yet there's no configure test 
> or automake conditional to keep people from accidentaly using an awk 
> that can't run jrtangle.  I can see why having texindex.awk in the 
> repository makes sense in that case.  But then the Makefile.am's have to 
> take that decision into account, too.

Sure.  I just don't think the easy way out of removing the files is
necessarily the best one.