Hello,<div><br></div><div>We observed a problem with savepoints in Mariadb 5.5.25.  The problem at the SQL layer is that a &quot;rollback to savepoint&quot; statement would get an error after the savepoint was created.  The problem was caused by the savepoint name in the savepoint stack getting overwritten by the handlerton savepoint_set method.  </div>
<div><br></div><div>Here is what we think is going on:</div><div>1. space is reserved in the savepoint memory for each handlerton in the ha_initialize_handlerton function.  </div><div>2. then space is reserved for the SAVEPOINT object in ha_init.  (Yes, ha_init gets called AFTER the ha_initialize_handlerton calls.</div>
<div>3. ha_savepoint computes the address of the handlerton&#39;s object as (uchar*)(sv+1) + ht-&gt;savepoint_offset.  this assumes that the SAVEPOINT object is laid out first, followed by the handlerton&#39;s space.  unfortunately, step (1) occurs before step (2) so this is not the case.</div>
<div><br></div><div>Here is a possible fix:</div><div>Move the base initialization of the savepoint_alloc_size from ha_init to the global intiializer.  With this change, the handlerton&#39;s offsets are computed AFTER the SAVEPOINT space is reserved. </div>
<div><br></div><div>Is this correct?</div><div><br></div><div>Thanks</div><div>Rich Prohaska</div><div><br></div>