On Mon, 11 Jun 2001, Tom Lane wrote:
> > It can't be resolved during the planning (directfunctioncall) because I do
> > want queries of a << b (b isn't a constant) to be also using the same
> > mechanism. (so far it looks like special_index_* can cope with that OK)
>
> You're mistaken ... that's not supported currently.
Augh, you are right. Well, now I have three options
a) fix match_clause_to_indexkey to use something like special_index_* for
'special' operators that can use indices in a nested loop-join.
b) add another access method to btree and muck with pg_amop (uuugh, bad
idea)
c) decide that I'm way deeper than I wanted to be already, and forget
about the idea.
Where I need a<<=b, I guess I can just use a>=network(b) and
a<next_network(b) directly.
Tom, thanks for helping me out.
I'll still follow up with set_masklen and network_text funcs.
-alex