Tom Lane <tgl@sss.pgh.pa.us> writes:
> > I suppose it makes more sense to optimize this on the basis of what's used in
> > the planner and executor rather than ALTER TABLE commands though.
>
> No, definitely not. Syscaches only exist to support hard-wired lookups
> in the backend C code. Indexes on system catalogs are of interest to
> the planner, but not syscaches. (So it is legitimate to have indexes
> with no associated syscache. The other way is not possible, though,
> because the syscache mechanism depends upon having a matching index.)
I imagine the planner and/or executor have to execute hard-wired lookups in C
code all day long to check for children of tables before they can execute
queries on those tables. I meant that the performance of those lookups was
undoubtedly more critical than the performance of DDL.
--
greg