Gerald Britton <gerald.britton@gmail.com> writes:
> Back to where I started in my top post: I became interested in this due to
> the doc note on returning a cursor and that it can be an efficient way to
> handle large result sets. I suppose that implies lazy evaluation. Does
> that mean that if I need plpgsql for a function for he language's power yet
> want the results to be returned lazily, a cursor is the (only?) way to go?
Nope. The docs' reference to a cursor only suggests that if you can
express the function's result as a single SQL query, then opening a
cursor for that query and returning the cursor name will work. But
if you need plpgsql to express the computation, that's not a terribly
helpful suggestion.
If you'd like to see some actual movement on the missing feature about
lazy evaluation in FROM, you could help test/review the pending patch
about it:
https://commitfest.postgresql.org/26/2372/
However, that still is only half of the problem, because you also need
a PL that is prepared to cooperate, which I don't believe plpgsql is.
I think (might be wrong) that a plpython function using "yield" can
be made to compute its results lazily.
regards, tom lane