An index-only scan fills its result slot from the HeapTuple an index
AM returns in scan->xs_hitup by deforming it with the virtual slot's
tuple descriptor. But the index AM formed that tuple with its own
descriptor, scan->xs_hitupdesc, and the two may disagree about tuple
layout. While each individual column's data *is* binary compatible,
that doesn't ensure that each "described tuple shape" has compatible
alignment.
This is a bug. All stable releases are affected.
Attached bugfix teaches nodeIndexonlyscan.c to deform using the
authoritative xs_hitupdesc descriptor used by the index AM, rather
than assuming that the descriptor used by the scan's virtual slot is
100% compatible with xs_hitupdesc.
The patch includes a test case demonstrating a query where this leads
to incorrect behavior when scanning a multicolumn GiST index. Without
the fix the executor gets confused about where the second index column
begins, leading to a spurious error (e.g., "ERROR: type with OID nnnnn
does not exist"). I haven't investigated whether the consequences
could be worse than just an error.
FWIW I discovered this bug when I wondered why the code in question
didn't already work this way. I was trying to simplify my own patch
that moves this functionality out of the core executor and into a
table AM helper function (needed for the index prefetching work). The
core executor already uses scandesc->xs_itupdesc for
scandesc->xs_itup, so why doesn't it also use scandesc->xs_hitupdesc
for scandesc->xs_hitup? I initially thought the bugfix's approach was
just simpler, but subsequently realized it was actually necessary for
correctness in certain edge cases.
--
Peter Geoghegan
