On Sat, 2005-09-17 at 16:04 -0400, David Walluck wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Thomas Fitzsimmons wrote:
>
> >>The post and postun problems I am having seem to be because
> >> /usr/bin/rebuild-gcj-db can not be found during post and postun. If
> >>Requires: java-gcj-compat >= 1.0.31 is used, whenever
> >> the post or postun is taking place, rebuild-gcj-db will be present.
> >
> >
> > I just changed rebuild-gcj-db to be an alternative symlink. Therefore
> > RPM wouldn't know about it. Might that be the problem?
>
> Is it even correct to have rebuild-gcj-db as an alternative? This file
> must exist always (even when not running gcj). Since the link is a slave
> to java, if the java alternative is changed (say, to a Sun 1.5.0 jvm),
> then I assume that this link will get lost? Or is that not true since
> gcj is the only one providing it?
>
> In any case, I have a similar problem of this link not being set up. And
> maybe since the Requires(post) is on the java-gcj-compat package and not
> %{_bindir}/rebuild-gcj-db itself, maybe rpm is not catching it.
I made rebuild-gcj-db an alternative for two reasons. One, so that two
java-gcj-compats could be installed in parallel and two because rebuild-
gcj-db references gcj-dbtool which is different for different gcj
installations (gcj-dbtool, gcj-dbtool4). The same arguments apply to
aot-compile-rpm.
You raise good points for why this isn't workable. One potential
solution is to only include these scripts in the "system default" java-
gcj-compat, since there will presumably only be one system-wide gcj
database on which aot-compile-rpm and rebuild-gcj-db should operate.
I'll fix this soon.
>
> Also, I think there's a packaging problem with java-gcj-compat not
> owning a lot of symlinks it creates. As I understand it, we can't own
> alternatives symlinks since this can cause conflicts between packages
> (yet why does rpm allow multiple owners of directories?). But there are
> other links created in post that probably could be %ghost'ed (the
> internal ones which link to libgcj.jar).
Yes, also a good point; I'll fix this too.
Thanks for the feedback,
Tom
