Since Warren changed RDF resource and datasource construction to route through the XPCOM repository, performance has slowed down significantly; e.g., it now takes ~10sec to read in bookmarks. What's so slow about the repository?

Reassigned to dp; From his email ---- The performance problem is because ProgID to CLSID conversion now hits the registry every time. I am changing the repository to use nsIRegistry and then will implement the cache of progid. That should get this back on track. ----

Checked in a patch that caches ProgID-to-CLSID mappings (including misses). This really needs to be re-worked: the code is starting to get ugly. :)

is the current behavior good enough for M3? or is someone looking at this now? lets move to M4 if performance is ok with the hack, and replace the hack in m4.

Fixed multiple loads. This should improve quite a bit. We can leave the rest for M4.

Bookmarks performance seems OK for M5. DP, could you check to make sure all performance enhancements are completed for M6?

Fine tuning is what is left.

This was more of a place holder for all the performance improvements we were planning on. They are all done.