Jesse Keating wrote:
On Wed, 2009-02-18 at 04:48 +0100, Ralf Corsepius wrote:
Well, there is one change in F11, which introduces a need for an least
partial ordered rebuild:
rpm has started to automatically add
Provides: pkgconfig(foo)
and to automatically add
Requires: pkgconfig(foo)
This requires an ordered rebuild in cases, when a package is part of a
chain of (devel) package deps.
Hrm, I think this would only matter if the rebuilt package that picks up
a Requires: pkgconfig(foo) made it into the repodata that was then used
by further builds, but before the packages that have Provides:
pkgconfig(foo) also made it into the repodata.
Agreed, however this already is the case.
When upgrading OSG in rawhide last weekend, I had encountered this issue.
When rebuilding a chain of package osgal, osgcal, poker3d, I had
intended to rebuild against the new OSG, had failed because
pkgconfig(cal3d) had not been provided.
I.e. I had to rebuild cal3d to let cal3d-devel provide it, before this
package chain could be rebuilt.
If we delayed tagging of
all of these builds until the very end, we could avoid the breakage.
Not sure this is completely desired though, but given the package set
where this would hit, I can picture lots of loops that we wouldn't be
able to fully order out.
AFAICT, the impact of this change can be reduced, by checking which
packages in current rawhide contain *.pc's but do not Provide nor
Require pkgconfig(foo) and to rebuild them (in manually presorted order)
in advance to the mass rebuild.
This all will likely get stuck if one of these packages doesn't survive
GCC44 (Which is what I expect to happen).
Starting to think that delaying the tag events is what we want here, get
the whole mess into the repodata at once rather than piecemeal as the
builds progress.
Ralf
--
fedora-devel-list mailing list
fedora-devel-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/fedora-devel-list
