[Yum-devel] [Patch] Resolver Performance and Correctness

[Jeremy Katz]

On Wed, 2007-06-06 at 12:08 +0200, Florian Festi wrote:
> Jeremy Katz wrote:
> > On Tue, 2007-06-05 at 13:22 +0200, Florian Festi wrote:
> >>    * added local rpms to pkgSack via localSack (cli.py)
> > 
> > Hrmm... what are you trying to accomplish with this?  Other API users
> > are almost certainly not doing this
>
> Reason for this is the way .whatInTsProvides/Requires are implemented. They 
> do no longer search within the TransactionInfo but search within the rpmSack 
> and the pkgSack and then filter the result to the pkgs really included in 
> the transaction - in a sense of all pkgs staying installed + all pkgs 
> getting installed. This allows to make use of the indices within those two 
> databases.
> 
> Command line pkgs don't fit into that scheme per se. To make them fit they 
> can be added to the pkgSack (with an PackageSack that creates its own 
> indices for fast search).  That way they are found and not filtered out as 
> long they are part of the transaction. This is exactly the wanted behavior 
> and doesn't need any additional code than adding them to the pkgSack.
> 
> I see the problem that other API users may not use the cli.localInstall but 
> just add the pkg objects to the transaction. We probably need to find a 
> solution at a lower level like checking on addition to the TransactionInfo 
> although I have to admit that I don't like this idea.

They're definitely not going to be using cli.localInstall as cli is
under /usr/share/yum.  If we go the above route, we're probably going to
have to do the addition when adding to the transaction set.  I don't see
any other common place that it could be

Jeremy