Re: Constraint stuff

On Mon, 7 Aug 2000, Tom Lane wrote:

> Jan Wieck <janwieck(at)Yahoo(dot)com> writes:
> > PL/Tcl and PL/pgSQL will load a function's source only once a
> > session. The functions loaded are identified by OID, so if
> > you drop/create a function, the PL handler will simply load a
> > different function too (from his point of view). At the time
> > we are able to ALTER a function, we might want to include
> > some version counter to pg_proc and in the fmgr_info?
>
> More generally, the PL functions need to be able to deal with
> recomputing saved plans after an ALTER of a table referenced by the
> function. I haven't really thought about how to do that ... but it

More and more generally, IMHO all saved plans need some validity
checking and not only for ALTER, but also for all operation those
changing relevant system tables. And this is not problem for PL only,
but for all what is based on SPI (and VIEWs?).

IMHO correct solution is _one_ space and one method for plans saving
= query/plan cache, and some common validity-checker that will work
over this cache. But how implement validity-checker is unknown...
(? Call from all command that changing system tables some handler, that
check plans in a cache ?)

Karel

BTW. - first step in this problem is (or can be) on the way ---
query cache...