2012/8/28 Jim Nasby <j...@nasby.net>: > On 8/28/12 2:51 PM, Pavel Stehule wrote: >>> >>> >The thing I don't like about this is it assumes that time is the best >>> > way to >>> >refer to when things changed in a system. Not only is that a bad >>> > assumption, >>> >it also means that relating things to history becomes messy. >> >> On second hand I don't have a problem with some optional counter, >> although I think so database system time is very useful and other >> counters for versioning are not necessary - because in one time I can >> have only one version - it doesn't do versions from rollbacked >> transactions. > > > What happens if the system clock runs backwards? >
probably, than you have more significant issues than this - it can be same with overloading any counter > What happens if two transactions start in the same microsecond? (And I know > for a fact that's possible, because I've seen it). yes, it is possible - and probably you need to know end of transaction - commit time - auxilary counter doesn't help - because it can be in different order too - when first transacttion was rollbacked Pavel > > More importantly, I believe using time to handle recording a versioned > history of something is flawed to begin with. You might care about what time > a new version was created; but what's far more important is recording the > correct ordering of things, and time isn't actually a great way to do that. > > Note that no version control systems use time as their primary attribute. > > -- > Jim C. Nasby, Database Architect j...@nasby.net > 512.569.9461 (cell) http://jim.nasby.net -- Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers
