%makeinstall vs DESTDIR (was: Re: rpms/poker-eval/FC-4 poker-eval.spec,1.8,1.9)

John Dennis jdennis at redhat.com
Tue Jun 20 03:43:23 UTC 2006


On Tue, 2006-06-20 at 05:11 +0200, Ralf Corsepius wrote:
> On Mon, 2006-06-19 at 19:26 +0300, Ville Skyttä wrote:
> > On Mon, 2006-06-19 at 08:59 +0100, Paul Howarth wrote:
> > 
> > > Actually the single best reason hasn't been included here.
> > [...]
> > 
> > Additionally, note that the DESTDIR approach is the documented way of
> > doing staged installs with automake and the Makefile Conventions section
> > of the GNU Coding Standards.  That in addition to personal experience
> > makes me think many upstreams are much more likely to pay attention to
> > having a working DESTDIR setup compared to the one %makeinstall
> > currently expands to,
> Well, all properly packaged automake-based projects support both styles,
> equally. "make distcheck" tests for both styles :)

Sorry, I missed the original thread, but I it is my belief %makeinstall
is fundamentally broken because of hardcoded assumptions. %makeinstall
is evil because it works under limited conditions which leads to the
mistaken belief it is a robust and proven mechanism which it is not.

"make install" with DESTDIR set to the buildroot is to the best of my
knowledge the most correct and robust idiom for populating the buildroot
for RPM's based on automake sources.

I can't comment on "distcheck", but %makeinstall should be avoided in
favor of DESTDIR installations IMHO. This is topical for me because I
just helped someone debug a spec file which was using %makeinstall and
was blowing up as it tried to install into the build machine's /usr
during the %install phase.

-- 
John Dennis <jdennis at redhat.com>




More information about the fedora-extras-list mailing list