I wrote:
> Hmm ... I agree that this is better normally. But there's an
> edge case where it would fail to notice a problem that the
> existing code does notice: if blocknum is close to UINT32_MAX
> and adding nblocks causes it to wrap around to a small value.
> Is there an inexpensive way to catch that?
After a few minutes' thought, how about:
Assert((uint64) blocknum + (uint64) nblocks <= (uint64) mdnblocks(reln, forknum));
This'd stop being helpful if we ever widen BlockNumber to 64 bits,
but I think that's unlikely. (Partitioning seems like a better answer
for giant tables.)
regards, tom lane