Christopher Kings-Lynne <chriskl@familyhealth.com.au> writes:
> Updated to match Tom's new semantics. passes my own regression tests plus
> all existing ones...
Checked and applied. I made one nontrivial change --- for anything on
which there's a syscache, it's a lot easier (and faster) to use
SearchSysCacheCopy to get a tuple to update than it is to do a heapscan.
The syscache will either have the tuple already, or will get it with an
indexscan --- either way wins over a heapscan when the catalog is at
all large.
(One of the things on my to-do list is to get rid of unnecessary
heapscans on system tables; there are still a few of 'em in code
paths that are used enough to be worth worrying about.)
regards, tom lane