On Feb 16, 2011, at 11:36 AM, Per Øyvind Karlsen wrote: >> > Would this patch be better? :) >
Yah, that looks better for performance and less of a radical change (so less risk) than trying to always attach an accurate count to every iterator. OTOH, performance and clarity are at odds here. There's more that could and should be done in the iterator code. I got most of the goop out (my target was 50% fewer LOC, I think I got to 45% before having to "Ship it!" a year ago. This patch SHOULD fix the python StopIteration exception. But there's likely a bit more to do with %trigger* handling, the rpmmiCount() there can't possibly be right with more than one trigger firing and the rpmmiPrune() count adjustment. Won't be hard to fix, but will need better test cases: the toy trigger packages in tests/* fire a single trigger which is also almost all of the known usage cases everywhere. In order to handle per-file pattern triggers for cache rebuilds there's almost certainly going to be additional complexity. OTOH, performance and clarity are at odds here. There's more that could and should be done in iterator code. BTW, wareham.jbj.org running 2010.0 got installed yesterday, moving the gatech tree I rsync'd this weekend, and I'm almost getting back to devel rather than sysadmin. 73 de Jeff
smime.p7s
Description: S/MIME cryptographic signature