While mucking around in src/backend/utils/time/tqual.c today, I noticed
the following comment attached to HeapTupleSatisfiesNow:
* mao says 17 march 1993: the tests in this routine are correct;* if you think they're not, you're wrong,
andyou should think* about it again. i know, it happened to me. we don't need to* check commit time against
thestart time of this transaction* because 2ph locking protects us from doing the wrong thing.* if you mess
aroundhere, you'll break serializability. the only* problem with this code is that it does the wrong thing for
system* catalog updates, because the catalogs aren't subject to 2ph, so* the serializability guarantees we
providedon't extend to xacts* that do catalog accesses. this is unfortunate, but not critical.
Much as I hate to disturb a comment just before its 19th birthday, the
bit about two-phase locking and serializability hasn't been correct
since around 1999 (when MVCC was added). :-)
Dan
--
Dan R. K. Ports MIT CSAIL http://drkp.net/
