Storage operations in Aggregator were always single-threaded and performed in the main GUI thread. Particularly, feed updates were also performed in the main thread, including fetching already-existent items from the database, looking for modified ones, updating modified items and inserting new ones. That resulted in a terrible overall performance during feed updates — one could tell when Aggregator decides to update its feeds just by feeling how sluggish and slow LeechCraft interface suddenly becomes for a dozen of seconds.
That's in the past now. Yesterday all that feed updating machinery was moved to a separate thread, and only parsing the downloaded feeds is done in the main GUI thread now, but I doubt that it is really worth moving that into a separate thread as well — XML parsing is fast as hell and is hardly the performance hotspot.
The threaded architecture allows moving arbitrary operations into the storage thread as long it makes sense from the UI point of view. For example, marking whole channels as read or unread was also moved into that separate thread, but it hardly makes sense to fetch items of the currently selected channel to that thread — user would wait for this operation to complete anyway, and it's fast enough for user to not start doing other stuff.
Threaded storage is already tested with SQLite and PostgreSQL on Linux and generally seems to work. It should work with any PostgreSQL/MySQL installation and with most modern SQLite installations. Though, if SQLite was compiled without threading support, Aggregator would misbehave, likely resulting in segfaults. In this case, file a bug to your distro's SQLite maintainers. And, of course, the usual warning: threaded storage is still quite experimental and not thoroughly tested, so bugs may happen.
This change only applies to the master branch: it won't be included in the upcoming 0.5 release.
