Re: CVS commit: pkgsrc/math/R

[David Holland <dholland-pkgchanges%netbsd.org@localhost>]

On Thu, Dec 20, 2012 at 11:22:44PM +0100, Joerg Sonnenberger wrote:
 > > >>> ===> Creating toolchain wrappers for R-2.15.1
 > > >>> ERROR: g95>=0.91 is not installed; can't buildlink files.
 > > >>> 
 > > >>> which isn't supposed to happen at all, but especially not with pbulk.
 > > >> 
 > > >> It means that some buildlink3.mk files pulls it in indirectly and the
 > > >> corresponding package itself doesn't depend on g95.
 > > > 
 > > > ...no it doesn't, it means that it thinks it confirmed g95>=0.91
 > > > during do-depends, but when it went to buildlink it, it wasn't
 > > > actually there. Normally this is caused by having an out-of-date build
 > > > dir containing e.g. g95-0.90, so when pkgsrc goes to install g95>=0.91
 > > > it actually installs 0.90 instead. But this should not be possible
 > > > with pbulk.
 > > 
 > > Is what you mean: it can happen building package P, if P depends only
 > > indirectly on Q but none of the direct depends of P cause Q to be
 > > installed? Because I don't think that's true; in that case it should
 > > attempt to build Q from ../../examples/Q inside pbulk and fail in
 > > depends.
 > 
 > A includes B/b3.mk. B/b3.mk includes C/b3.mk, but B itself doesn't.
 > C is an indirect dependency of A, which will not be installed during
 > bulk builds at this point unless something else pulled it in.

Right, but that can't itself cause the observed behavior. If it's just
not installed, it should fail in the depends phase.

however, corrupting the buildlink tree can cause who-knows-what to
happen, so it's probably a good thing I went and tracked it down.

-- 
David A. Holland
dholland%netbsd.org@localhost