Re: [Tracker] Merging the indexer-split branch to TRUNK

["Michael Biebl" <mbiebl gmail com>]

2008/9/25 Martyn Russell <martyn imendio com>:
> Michael Biebl wrote:
> > 2008/9/25 Michael Biebl <mbiebl gmail com>:
> > > 2008/9/25 Martyn Russell <martyn imendio com>:
> > > > Michael Biebl wrote:
> > > > > 2008/9/25 Martyn Russell <martyn imendio com>:
> > > > > > Michael Biebl wrote:
> > > > > > > 2008/9/25 Martyn Russell <martyn imendio com>:
> > > > > > > > Jamie McCracken wrote:
> > > > > > > > > ok lets just have a shell script that deletes old stuff (which is not
> > > > > > > > > called by Makefile) and put a note in the README file on how to upgrade
> > > > > > > > Really, this is not necessary. It is only old binary files and old
> > > > > > > > application data which is causing the problem here. It isn't *user* data
> > > > > > > > at all. Removing this stuff when we install is perfectly fine.
> > > > > > > Let the user decide, but don't delete stuff in the system directory
> > > > > > > during compilation.
> > > > > > This doesn't happen during compilation, it happens during the install
> > > > > > process and the user might not know about about the choice in the first
> > > > > > place, all they see is it doesn't work.
> > > > > Yeah, I saw that you just committed that.
> > > > > Nonetheless, you are talking about developers or at least advanced users here.
> > > > >
> > > > > Who else would svn checkout and test compile tracker on his own, honestly.
> > > > The same people that will write to the mailing list asking why svn up
> > > > didn't work.
> > > >
> > > > > > This data is data we installed previously for the user. It is data we
> > > > > > would *uninstall* if the user ran make uninstall before they did svn up.
> > > > > Have you considered that this makes it impossible for me to have
> > > > > tracker 0.6.6 installed while trying to package 0.7?
> > > > >
> > > > > I don't want it to silently  nuke my perfectly working 0.6.6 setup.
> > > > Not really. Make distcheck should install to a specific prefix, not the
> > > > same one you are using for older tracker installs. Building debian
> > > > packages does the same. If you are using the same prefix for your real
> > > > applications as your packaged applications then I would say you're doing
> > > > something wrong there.
> > > Do you want to tell me how to do packaging? Ever heard of DESTDIR?
> >
> > Let me rephrase that: I you insist on keeping the current behaviour,
> > at the very least make it respect DESTDIR.
>
> I am trying to see how we don't do that now. We use $bindir, $libdir,
> and all the other Makefile.am variables in other places too and they
> work, why would it be any different here?

Please see [1] for further info on DESTDIR
Proper DESTDIR support is required, to build the Debian/Ubuntu packages.
The install stage runs unprivileged and usurally installes the files
into a tmp dir
debian/tmp via "make install DESTDIR=debian/tmp" from where the
package is created.

Afaik

So, to support DESTDIR, Makefile.am should look something like:

install-exec-hook:
        rm -Rf $(DESTDIR)$(bindir)/trackerd
        rm -Rf $(DESTDIR)$(bindir)/tracker-indexer
...

> But either way, I too agree with you in whole. Normally I wouldn't have
> added this. I simply added it because Jamie and I discussed it and it
> was symptomatic of what Jamie found when he upgraded and we agreed we
> would do something to fix it.
>
> Carlos suggested perhaps checking in configure.ac and putting it right
> at the bottom in BOLD letters so people know it will happen? So maybe
> that's better, then people are informed and know make install will
> replace old files. We could even add a command line switch for it to
> disable it, but I think that is going too far.

Yeah, that sounds overly complicated.

Michael

[1] http://www.gnu.org/software/libtool/manual/automake/DESTDIR.html
-- 
Why is it that all of the instruments seeking intelligent life in the
universe are pointed away from Earth?