bzr commit into mysql-5.4 branch (alik:2873) Bug#46650

[Alexander Nozdrin]

#At file:///mnt/raid/alik/MySQL/bzr/qqq/mysql-trunk/ based on revid:alik@stripped

 2873 Alexander Nozdrin	2009-08-24
      Backporting patch for Bug#46650 (Innodb assertion autoinc_lock == lock
      in lock_table_remove_low on INSERT SELECT) to mysql-trunk.

    modified:
      storage/innobase/lock/lock0lock.c
=== modified file 'storage/innobase/lock/lock0lock.c'
--- a/storage/innobase/lock/lock0lock.c	2009-07-30 12:42:56 +0000
+++ b/storage/innobase/lock/lock0lock.c	2009-08-24 06:48:57 +0000
@@ -3574,7 +3574,8 @@ lock_table_remove_low(
 		and lock_grant()). Therefore it can be empty and we
 		need to check for that. */
 
-		if (!ib_vector_is_empty(trx->autoinc_locks)) {
+		if (!lock_get_wait(lock)
+		    && !ib_vector_is_empty(trx->autoinc_locks)) {
 			lock_t*	autoinc_lock;
 
 			autoinc_lock = ib_vector_pop(trx->autoinc_locks);
@@ -3647,8 +3648,10 @@ lock_table_enqueue_waiting(
 
 	if (lock_deadlock_occurs(lock, trx)) {
 
-		lock_reset_lock_and_trx_wait(lock);
+		/* The order here is important, we don't want to
+		lose the state of the lock before calling remove. */
 		lock_table_remove_low(lock);
+		lock_reset_lock_and_trx_wait(lock);
 
 		return(DB_DEADLOCK);
 	}


Attachment: [text/bzr-bundle] bzr/[e-mail]