On Sat, Dec 10, 2011 at 09:28:03PM +1000, Allan McRae wrote:
> On 08/12/11 02:49, Lukas Fleischer wrote:
> > On Tue, Dec 06, 2011 at 05:05:20PM -0600, Dan McGee wrote:
> >> On Tue, Dec 6, 2011 at 4:58 PM, Lukas Fleischer
> >> <firstname.lastname@example.org> wrote:
> >>> On Tue, Dec 06, 2011 at 04:38:41PM -0600, Dan McGee wrote:
> >>>> On Tue, Dec 6, 2011 at 3:29 PM, Lukas Fleischer
> >>>> <email@example.com> wrote:
> >>>>> Can some autotools wizard please check if I broke anything here? `make
> >>>>> distcheck` still looks good to me but I'm not 100% sure what I broke by
> >>>>> adding all scripts to "bin_SCRIPTS"
> >>>> Before: `make install` does not install contrib/.
> >>>> After: `make install` installs contrib/. So probably not what we want,
> >>>> at least recursively or by default. It would be nice that if you were
> >>>> actually in contrib, `make install` would do what one expected (and
> >>>> kill the ugly hardcoding of each script needed to currently install
> >>>> ). Autotools probably makes a setup like this a bitch though.
> >>>> Thoughts from anyone else?
> >>> How about removing "contrib" from "SUBDIRS" in our main Makefile.am and
> >>> adding something like "DIST_SUBDIRS = $(SUBDIRS) contrib"? This way,
> >>> contrib scripts won't be installed when running `make install` in the
> >>> top level source directory but can be easily installed by running `make
> >>> install` in "contrib/"...
> >> The problem there is it completely stops make from descending in
> >> there, and as I've seen many times before (namely with docs), it gets
> >> ugly when I have to package it. I think you'd be out of luck because
> >> your Makefile wouldn't even be generated in there, and thus *.in never
> >> gets converted to the scripts and other files.
> > It doesn't completely stop make from descending in there. As discussed
> > on IRC, directories listed in "DIST_SUBDIRS" will still be honored when
> > running `make dist`. The only difference is that we would have to run
> > `make` and `make install` separately in "contrib/". This makes sense to
> > me because:
> > * I consider contrib scripts a sub-project of (and not part of) pacman.
> > If you say that they actually do belong to pacman, there should be a
> > possibility to install contrib scripts when using make(1) (e.g. by
> > introducing a configure option like "--with-contrib"). This seems very
> > natural, at least.
> > Given that we probably do not want to install contrib scripts together
> > with pacman and prefer separate install trees/procedures, it doesn't
> > make any sense to build contrib scripts in the same procedure we use
> > for building pacman, also. Basically, saying that you will have to
> > run `make install` separately since it's a different sub-project but
> > build it when building the main project feels a bit obscure...
> > tl;dr: To me, only one of these options make sense:
> > - Make contrib scripts optional but build *and* install them together
> > with pacman (that means "introduce a configure option").
> > - Use a different build *and* install procedure for contrib scripts
> > (that means "use DIST_SUBDIRS"!).
> This option makes sense to me (and I have seen that sort of idea used
> One query about this approach... Will running ./configure still
> generate the Makefile in contrib/ when doing this? Because having to
> maintain a separate configure script that does the same substitutions as
> the main one would be annoying.
Yeah, it will still generate the Makefile. I tested this running
`./autoclean.sh; ./autogen.sh; ./configure`.