For performance reasons we only permit the file: operator in the
watched project filters. We have the data on hand in memory for
formatting in the message, so its reasonably cheap to scan through
filters using the file operator. But doing it against a database
query is just too slow, odds are the file list isn't in the diff
cache and thus has to be computed on the fly. At 200ms up to as many
as 4 seconds per change, this is simply unacceptable performance.
Bug: issue 70
Change-Id: I15ceb262419132a32c828e4051853b15ab963876
Signed-off-by: Shawn O. Pearce <sop@google.com>