Re: Windowing Function Patch Review -> Standard Conformance

I wrote:
> I've been hacking on this and I have a grammar that pretty much works,
> but there's some bizarreness around UNBOUNDED. I'll post it later.

Here is a proof-of-concept grammar patch that allows frame_bound to use
a_expr instead of a hacked-up constant production (which, as I
complained before, didn't allow all the cases demanded by the spec).
The key changes involved are:

* BETWEEN has to become a fully reserved word instead of type_func_name.

* PARTITION RANGE ROWS FOLLOWING PRECEDING have to be assigned the same
explicit precedence as IDENT. (This doesn't have any bad consequences
AFAIK, except that you can't use a postfix operator in frame_bound exprs
unless you parenthesize it.)

* UNBOUNDED has to be assigned a precedence slightly lower than
PRECEDING and FOLLOWING, else it's ambiguous whether "UNBOUNDED
PRECEDING" ought to be parsed as "a_expr PRECEDING". I'm a bit nervous
about this solution; we might someday need to make some of these
keywords at least partly reserved instead. But right now it doesn't
seem to have any negative consequences.

Making BETWEEN fully reserved is a bit annoying, but it would only
break apps using BETWEEN as a type or function name. The former doesn't
seem like a problem, the latter might be.

Comments?

regards, tom lane

PS: I've removed some uninteresting hunks to shorten the diff, so you might
get some chatter from patch if you try to apply the diff.