I wrote:
> Hmm. It would actually work if we made CONCURRENTLY reserved instead;
> and that would fix Simon's gripe too. That's kind of weird from a
> standards-compliance POV, but in terms of the risk of breaking
> applications it might be better than reserving ON.
Wait a minute. I must have been looking at the wrong keyword list
--- ON already is reserved. The problem is exactly that it can't
tell whether CREATE INDEX CONCURRENTLY ON ... means to default the
index name or to create an index named CONCURRENTLY. So really the
*only* way to fix this is to make CONCURRENTLY be at least
type_func_name_keyword.
regards, tom lane