> These pointers are useless
> except in the very unusual case where one steps forward and then back
> in a sequential scan (for example, "FETCH 1; FETCH BACKWARD 1;" in a
> cursor).
Actually I think it should be known whether a fetch backward is actually
allowed for a particular cursor, since it usually needs to be declared
as a scroll cursor explicitly.
> It seems to me that this is wrongheaded. We could simplify
> and speed up the normal case by maintaining only a "current" pointer,
> which would be well worth the extra work in the forward/back
> step case.
So it probably does pay off for a cursor, that is explicitly requested
to be scrollable, but certainly not for others, yes.
Andreas