Hi,
I've pushed the fix, after a a couple extra rounds of careful testing.
I noticed that the existing pg_visibility regression tests don't check
if we freeze the TOAST rows too (failing to do that was one of the
symptoms). It'd be good to add that, because that would fail even
without CLOBBER_CACHE_ALWAYS, so attached is a test I propose to add.
regards
--
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company