Josh Berkus <josh@agliodbs.com> writes:
> yeah, you're right .. that would give us a short list of conversions
> which don't require a rewrite. However, as Tom points out, that
> doesn't mean that they might not need a reindex (as well as OID, there's
> also XML).
Um. I had actually forgotten about the reindexing point, but yup that
is a stumbling block to any "no work" conversions. It might be best to
only handle cases where the column's base type is not changing, so that
we don't have any index semantics changes happening. I think we could
still handle the varchar->text case (since they share index opclasses)
but that could be a hardwired special case.
regards, tom lane