Re: pg15b1: FailedAssertion("val > base", File: "...src/include/utils/relptr.h", Line: 67, PID: 30485)

On Tue, May 31, 2022 at 5:52 PM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Thomas Munro <thomas(dot)munro(at)gmail(dot)com> writes:
> > Count we make the relptrs 1-based, so that 0 is reserved as a sentinel
> > that has the nice memset(0) property?
>
> Hm ... almost. A +1 offset would mean that zero is ambiguous with a
> pointer to the byte just before the relptr. Maybe that case never
> arises in practice, but now that we've seen this problem I'm not real
> comfortable with such an assumption. But how about a -1 offset?
> Then zero would be ambiguous with a pointer to the second byte of the
> relptr, and I think I *am* prepared to assume that that has no use-cases.
>
> The other advantage of such a definition is that it'd help flush out
> anybody breaking the relptr abstraction ;-)

Seems backwards to me. A relative pointer is supposed to point to
something inside some range of memory, like a DSM gment -- it can
never be legally used to point to anything outside that segment. So it
seems to me that you could perfectly legally point to the second byte
of the segment, but never to the -1'th byte.

--
Robert Haas
EDB: http://www.enterprisedb.com