On Thu, Aug 11, 2016 at 5:24 PM, Robert Haas <robertmhaas@gmail.com> wrote:
> You argue against
> these things on the grounds that they might change later, but the
> overwhelming evidence from posts on this list is that people would
> prefer to have access to APIs that might not be stable rather than
> have no access at all.
I don't think it's useful to take this ultimatum approach. I would say
abstraction boundaries are a fairly well-proven C.S. tool at this
point -- and indeed by sometime last century. The real question is
where do the benefits outweigh the costs and that's going to be a
question of balancing conflicting priorities. Not one where an
ultimatum is justified.
So the real question is, are index access methods a place where we
want to take short cuts and just expose internals or is this a place
where we should spend the effort to design good abstractions? At face
value it certainly seems like a line worth defending but historically
it's been a kind of half-hearted abstraction since it was never clear
what new access methods might need and how to abstractly define every
possible attribute they might have. And the push away from SQL defined
attributes seems to be conceding defeat on that front.
--
greg