I wrote:
> In general we do check incoming binary values to ensure minimal
> validity. I think when I did timestamp_recv I was thinking it was
> just like int8 or float8 (respectively), in that any bit pattern is
> potentially legal; I had forgotten about the range restrictions.
> I haven't looked at the details of Stephen's patch (and can't till the
> archives site comes back up) but the idea is probably sound.
Having looked at it, I don't like the patch as-is; it misses
timestamptz_recv and doesn't handle the boundary condition
correctly for the HasCTZSet case. However the details of the latter
are likely to change completely once we finish adopting src/timezone.
I'll make a note to do something with this issue after the TZ patch
is in.
regards, tom lane