[Rpm-metadata] question about PreReq deprication...

Jeff Johnson jbj at redhat.com
Thu Oct 9 00:03:26 UTC 2003


On Wed, Oct 08, 2003 at 03:57:02PM -0500, Mark Hatle wrote:
> Jeff Johnson wrote:
> > Not any more. PreReq: just set a flag as a hint to the original (and 
> > broken!) topological
> > sort.
> > 
> > All releases of rpm since rpm-4.0 have a correct topological sort that 
> > continues to
> > provide the original guarantee of PreReq:
> >     A PreReq:'d dependency is guarantedd to be installed before the 
> > package that contains
> >     The PreReq:.
> > 
> > Since all Requires: are now guaranteed to be installed before each and 
> > every package
> > that contains (except in the face of dependency loops where topological 
> > sort is impoossible,
> > can't happen by definition or ordering), PreReq: is just a hysterical 
> > artifact, is completely
> > synonymous with Requires:.
> 
> During an initial software install, many times circular dependency loops 
>   can happen.. Does the use of PreReq help deal with this, or not?
> 

Not any more. What replaces is context markers, and an explicit algorithm
for unloopling loops, basically by ignoring a certain dpendency passed
from configuration, or by ignoring all dependencies in loop, but preserving
the PreReq: guarantee just in case it matters.

PreReq:'s don't matter, loops is loops, if you want order through topological
sort, then get rid of the loops.

> I am more then willing to not use the PreReq, but my understanding was 
> that it was there to handle the situation when a dependency loop occurs 
> during a large group of packages.

Not any more, install/erase contexts, or dependency whiteout, are used
by rpm to sort out loop problems.

73 de Jeff

-- 
Jeff Johnson	ARS N3NPQ
jbj at redhat.com (jbj at jbj.org)
Chapel Hill, NC



More information about the Rpm-metadata mailing list