On Jan 13, 2011, at 10:04 AM, Matthew Dawkins wrote:
> You know what I'd like to see is a list of the ordering done by rpm.org and the ordering done by rpm5. In my past experience going thru and eliminating the loops and this very same procedure with Jeff, made me realize that Mandriva packages have a weird looping of dependencies defined in the specfile Requires: among the basesystem-minimal packages. And many go along the lines A->B->C->A just because of say "one binary" or "config file" that needs to be in place before the actual rpm gets installed. And I think you began to see those errors.
A topological sort (with LOOP's) is what is done. Both the Tarjan SCCS algorithm
from @rpm.org as well as what RPM has always done are present.
Recent changes @rpm.org have added a Newer! Better! Bestest! ordering-only
requirement for constructing an initrd (no other usage case has been reported
Yes, Mandriva has weird looping issues. The dependencies that are "weird" are used
consistently enough that the loops can be detected a priori if necessary.
> So what I guess I'm trying to say is I am surprised that rpm.org got around this without those install errors.
> Two scenarios that I could see possibly different with rpm.org to rpm5 are:
The answer is this:
rpm.org removes all loops initially.
There is no getting around dependency loops.
> Rpm.org knows how to order when a binary or config file is needed to be in place before a package is installed to avoid rpm installation errors.
> Or the other scenario that I could believe is the rpm.org unpackages the contents of all the packages in a loop and then processes the pre, prein, post...etc.
> The question I have for both Jeff and Per, is flattening initial rpms packages to get these needed pieces in place before and actual chroot install begins and super unacceptable hack?
Meanwhile I can likely help getting your UL boot-strapped w/o using rpm2cpio.
73 de Jeff
Received on Thu Jan 13 17:03:34 2011
- application/pkcs7-signature attachment: smime.p7s