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

[Florian Festi]

Jeremy Katz wrote:
> On Wed, 2007-06-06 at 11:11 +0200, Florian Festi wrote:
>> Jeremy Katz wrote:
>>> On Tue, 2007-06-05 at 13:22 +0200, Florian Festi wrote:
>>>>    * TransactionData.relatedto and .depends_on converted to sets
>>> Should be fine.  This is the sort of thing that becomes a lot easier to
>>> integrate with separated out patchsets :)
>> Ok, this also contains a small speed up for .getMembers() but it should be a 
>> bit smaller than the original patch...
> 
> Applied the getMembers() change.  After thinking a little more, I wonder
> if we want to define an AppendableSet and use that for depends_on and
> related_to.  At the same time, I can't find anything directly modifying
> them so we're probably okay avoiding that overhead.  Any other opinions?
> 
> Jeremy

The code looks quite unfinished (removes of transaction members are not 
honored, not all relations are added) so I doubt it is used anywhere (except 
in the SortableTransactionData). I also don't see any reason why someone 
else than the resolver should add relations and why they shouldn't use 
.setAsDep().

The change doesn't affect the performance of smaller operations at all. But 
for an full install the lists use up 400 of 1000 seconds IIRC.

Florian