Replication aborts with ER_VIEW_SELECT_TMPTABLE after half-failed RENAME

--source include/have_binlog_format_mixed.inc

--source include/master-slave.inc

CREATE TABLE t1 (a INT);

CREATE TEMPORARY TABLE t1 (b INT);

--error 0,ER_CANT_REOPEN_TABLE,ER_TABLE_EXISTS_ERROR

RENAME TABLE t1 TO tmp, tmp TO t1;

CREATE VIEW v AS SELECT * FROM t1;

--sync_slave_with_master

# Cleanup

--connection master

DROP VIEW v;

DROP TABLE t1;

--source include/rpl_end.inc

In the test case above, RENAME TABLE behaves differently on different versions.

• on 5.5 it succeeds, and t1 ends up as a temporary table. Thus, the next CREATE VIEW already fails on the master, and there is no discrepancy with the slave;
• on 10.0 and 10.1 it fails completely with ER_CANT_REOPEN_TABLE, so t1 remains being a temporary table, thus the next CREATE VIEW also fails on the master, and there is no discrepancy on the slave;
• but on 10.2 and 10.3, it fails partially with ER_TABLE_EXISTS_ERROR – first, it renames the temporary table t1 to tmp}, and then it fails to rename {{tmp to t1, so it ends up with t1 being a base table, and the next CREATE VIEW succeeds on the master; but it doesn't write the half-failed RENAME to the binary log, so the slave still have t1 as a temporary table, and CREATE VIEW fails there with

  Last_SQL_Error Error 'View's SELECT refers to a temporary table 't1'' on query. Default database: 'test'. Query: 'CREATE ALGORITHM=UNDEFINED DEFINER=`root`@`localhost` SQL SECURITY DEFINER VIEW `v` AS SELECT * FROM t1'

I'm not quite sure which part is a bug – maybe it's wrong that the second part of RENAME fails, as it attempts to rename tmp to a base table rather than temporary table; or maybe it should be so, but then it should be written to the binary log (with the error code). In any case, something is wrong, since it causes a replication abort.