MDEV-15395 Wrong result or Assertion `old_part_id == m_last_part' fai…

…led in ha_partition::update_row on slave

RBR cannot work with system versioning on the master.

row_end column is either system time (not @@timestamp) with microsecond
precision or transaction id. Either way, it'll certainly be different
on the slave. So if the master row contains row_end column, it won't
match on the slave. And if we ignore row_end when comparing,
then some other row might match instead.

mysql-test/suite/versioning/t/rpl.test

@@ -1,18 +1,6 @@
---source include/master-slave.inc
-
-if ($MTR_COMBINATION_STMT)
-{
-  --source include/have_binlog_format_statement.inc
-}
-if ($MTR_COMBINATION_ROW)
-{
-  --source include/have_binlog_format_row.inc
-}
-if ($MTR_COMBINATION_MIX)
-{
-  --source include/have_binlog_format_mixed.inc
-}
+--source include/have_binlog_format_mixed_or_statement.inc
 --source suite/versioning/engines.inc
+--source include/master-slave.inc
 
 #BUG#12662190 - COM_COMMIT IS NOT INCREMENTED FROM THE BINARY LOGS ON SLAVE, COM_BEGIN IS
 #Testing command counters -BEFORE.

sql/sql_class.cc

@@ -6197,6 +6197,9 @@ int THD::decide_logging_format(TABLE_LIST *tables)
         flags_write_some_set |= flags;
         is_write= TRUE;
 
+        if (table->table->versioned())
+          flags_write_all_set &= HA_BINLOG_STMT_CAPABLE;
+
         prev_write_table= table->table;
 
       }