Hi, I have been simulating a crash during a transaction. After BEGIN, at some point I do exit(-1) instead of COMMIT or ROLLBACK in order to simulate a crash.
After that, I can see that a -journal file is present. If I restart my executable, it seems that the changes of the transaction are made undone (which is as expected). The journal file however is not removed. Furthermore, if I try to do the same operation again (BEGIN + some changes), I get an SQL_BUSY error code on the first record change (UPDATE/SET). I have the feeling that the OS still has a lock on the database. Any ideas on how to prevent this or on how to recover from this situation? I am using the following setup: -sqlite-3.2.1 -linux/i686/2.6.9-1.667smp -application using posix threads. Only one thread is accessing the database. Thanks, J.J. Krabbendam
