On Fri, Feb 18, 2011 at 11:38, Julian Foad <julian.f...@wandisco.com> wrote: > On Fri, 2011-02-18 at 17:20 +0100, Branko Čibej wrote: >... >> If I read that patch correctly, you expect to use one sqlite transaction >> per pristine file? While this may not affect WC performance now, it'll >> surely show up once other queries are optimized and transactions made a >> bit less granular. > > Yup - I mentioned that earlier today in the other thread "Pristine store > - spec": I think I'll need to extend so that a buch of pristines can be > installed at once. Glad to find you're thinking the same. But I'll get > the concurrency guarantees sorted out before extending it.
Let's get the functionality correct, and then worry about the lifetimes of sqlite transactions. I mean... how far behind are we? If we're not terribly behind, then I would suggest publishing 1.7 as-is. Then we can look at a larger body of work to reduce the sqlite txn's in the 1.8 timeframe. (eg. some lower-level APIs that auto-bulk changes into a large txn). Can somebody provide a pointer to some of the latest speed analysis? Thanks, -g