Neil Conway <neilc(at)samurai(dot)com> writes:
> On Fri, 2003-03-21 at 12:12, Tom Lane wrote:
>> (No problem here with adding the noise-word option, of course.)
> Note that it won't be a noise word: if NO SCROLL is specified, an
> attempt to do a backward fetch on a non-scrollable cursor will yield an
> error.
Hm. We could imagine a three-way option:
NO SCROLL always error if fetch backwards
<nothing> allow if plan supports it
SCROLL allow, add overhead if needed to support
The <nothing> case would deviate from the spec but would give backwards
compatibility.
regards, tom lane