On tis, 2011-03-15 at 16:16 -0400, Tom Lane wrote:
> But what did you think of the idea of setting collations during a
> post-pass, so that the collation derivation values need only be local
> storage during that one recursive routine?
That sounds reasonable. We do need the collation value in the
expression tree permanently, but not the collation derivation, as far as
I can tell.