On Fri, Apr 04, 2003 at 11:19:43AM -0500, Matt Wilson wrote:
> On Fri, Apr 04, 2003 at 10:18:39AM -0500, Bill Rugolsky Jr. wrote:
> >
> > When you say "not enabled" do you mean that the distro itself was
> > not prelinked in the distributed RPMS? Is that the intention for
> > future releases -- to distribute the executable code in the RPMS
> > prelinked so it "just works"?
>
> The intention is to have a cron job do the prelinking in future
> releases. There's no way to prelink inside the RPMS themselves.
I was thinking of "rpmbuild -bi" the whole distro into a BuildRoot,
prelink the whole thing, and then rpmbuild --short-circuit -ba
(assuming one has the Mandrake short-circuit patch ;-)). This does
introduce a "reproducible build" problem, but that doesn't seem
insurmountable. Actually, I've been contemplating post-processing
the binary RPMS in RHL 9 -- unpack it, prelink it, repack the files
into prelinked RPMS.
One of my concerns is about the administrative and security
implications of prelinking. On our systems, we do nightly checksum
comparisons laterally (across nearly identically configured systems)
and longitudinally (across time).
RPM handles this situation with prelink undo, but it's ugly IMHO. Also,
say I want to mount /usr or /opt read-only, possibly shared from an NFS
server. Without incremental prelinking, and given the the practical need
to use "prelink -m" on 32-bit platforms, the world is messy. On 64-bit
platforms incrementally assigning deterministic addresses with plenty of
space between them should not be an issue, so that should allow things
to be locked down once configured. I'd like to be able to prelink the
stuff on /opt once and forget about it.
[Even without prelinking, RPM opens a whole can of worms in that
situation, such as how do I deal with "extra" NFS shared packages (e.g.,
OpenDX) that depend on locally installed resources. Do I replicate
that information in RPM databases across all machines, or only do
RPM dependency tracking on a "master" host?]
> > In RHL 8.0, it was taking me about 25 seconds to re-login into GNOME
> > (i.e., with a warm cache) on an 800 MHz Celeron. So I prelinked, but
> > noted that there were lots of GNOME-related lib warnings about implicit
> > dependencies, so I didn't expect much speed up. Nevertheless, this
> > reduced re-login to 6-7 seconds.
>
> What kind of warnings?
undefined non-weak symbols
Regards,
Bill Rugolsky
