-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On 07/08/2015 10:33 AM, Brian Soby wrote: > Nothing appears to happen but it's not running very faithfully. > Specifically, I get a bunch of problems on the rollbacks about the > savepoints not existing even though it's replaying the original > SAVEPOINT "x" command before the rollbacks. This is probably more > due to me not knowing what's going on than anything else.
Do you have pysqlite configured with isolation level of None? If you use the defaults then it starts doing stuff on your behalf that you wouldn't have captured. https://github.com/ghaering/pysqlite/blob/51c1a18eac3d1d14a052b09e3546dd0c4ce70638/src/cursor.c#L607 Ultimately since you understand your code base best, you'll need to either provide pure SQL that can reproduce the issue, or you'll need to prove that there is no other component in the process that is corrupting memory that happens to be used by SQLite. To better catch issues with the SQLite shell, I recommend compiling it with -DSQLITE_DEBUG -UNDEBUG and running under valgrind. This will improve the chances of catching memory issues and verify SQLite internal assertion checking (there is a lot of it). Roger -----BEGIN PGP SIGNATURE----- Version: GnuPG v1 iEYEARECAAYFAlWet/wACgkQmOOfHg372QT+ggCgwXiWdIhegZlUdDINi9OhN4G6 5gwAn22bk2Qgt2ljVHb5c7GBSYFkALAK =5mrB -----END PGP SIGNATURE-----