I discovered while fooling around the other night that the named
restore point patch introduced a small bug into recoveryStopsHere():
the test at the top of the function now lets through two
resource-manager IDs rather than one, but the remainder of the
function tests only the record_info flag and not the
resource-manager-id. So the test for record_info == XLOG_XACT_COMMIT,
for example, will also return true for an XLOG_CHECKPOINT_SHUTDOWN
record, but the decoded commit time will be some random garbage rather
than a commit time, because the format of the record is totally
different.
Absent objections, I'll push the attached fix.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company