Re: Lookup penalty for VARIADIC patch

On Jul 15, 2008, at 4:58 PM, Tom Lane wrote:
> There are two ways we
> could do it: a bool column that is TRUE if the function is variadic,
> or an oid column that is the variadic array's element type, or zero
> if the function isn't variadic. The second would take more space but
> would avoid having to do a catalog lookup to get the element type in
> the case that the function is indeed variadic. I'm leaning to the
> second way but wanted to know if anyone objected?

If you go the second route, I'd vote for it being NULL if the
function isn't variadic, unless that would play hell with the C side
of the catalog code...

> Also, it occurs to me that we could buy back a good part of the extra
> space if we allowed pg_proc.probin to be NULL for internal functions.
> Right now it's always "-" in that case, which is useless ...

I'd vote for that being NULL in any case... magic values should be
avoided when possible.
--
Decibel!, aka Jim C. Nasby, Database Architect decibel(at)decibel(dot)org
Give your computer some brain candy! www.distributed.net Team #1828