> I can imagine a protocol that will improve performance by keeping track > of whether the database was updated, which would allow a quick > determination whether a program's cache was still valid. The idea is to > increment a 'version number' in the database file as soon as a program > obtains a write lock. A reader could then determine whether the file was > written to and flush the cache only when necessary, speeding things up a > little. A 128 bit counter should never wrap around.
Such a counter (32-bits only) was included in the file-format for version 3. Current versions of the library update the counter every commit, but it's not being used to allow persistent caches yet. I think such a change would require a lot of testing before it would be safe enough to use. And as you say, sophisticated operating systems do a pretty good job of caching the file anyway. __________________________________ Discover Yahoo! Find restaurants, movies, travel and more fun for the weekend. Check it out! http://discover.yahoo.com/weekend.html