How do I cancel an operation in Pro*C (with threads)

Question

I would like to be able to cancel a potentially long running Oracle operation initiated in Pro*C. It seems that, in an application having only a single database connection (i.e. does not use EXEC SQL ENABLE THREADS), I can do this by sending the SIGINT signal from another thread -- since obviously the DB thread is blocked waiting for the operation to complete. By having the "cancel thread" execute a pthread_kill(tid, SIGINT) against the blocked thread the operation is cancelled and the requisite Oracle error (ORA-01013) is caught.
However, it seems that once I enable threads, this method stops working. Unfortunately, the application I am writing needs multiple database threads. More dismal, I need to direct the cancellation to a single thread. I can, of course, direct the SIGINT to a thread via pthread_kill, but it does not work. Even a single thread with a single DB connection ignores the attempt to cancel -- in fact the program I wrote to test my initial theory stopped working (well, stopped canceling the operation) immediately just by adding an EXEC SQL ENABLE THREADS call.
So...does anyone know a call that I can make, from within another thread, that would cause an operation (executing on a given database context (sql_context) to become cancelled? Or does anyone have any other suggestions that I may experiment with?
Thanks all! Looking forward to any ideas.
OS: AIX 5.3
DB: Oracle 10g

Answer 1

Well, while I am still unsure why enabling threads renders my original solution incapable, I have found a suitable working solution.
After some research, I came across the OCIBreak call in the Oracle Call Interface. Mating this with the Pro*C functions to get the environment and service contexts (SQLEnvGet and SQLSvcCtxGet) allows me to call it. So, currently I set a SIGALRM signal handler and use pthread_kill in the monitoring thread to fire it off. The signal handler eventually calls OCIBreak.
I am hoping I can get rid of the signal handler and directly call OCIBreak from within the the monitoring thread itself -- but I'm not sure if this will break Oracle's thread consistency model and cause any heart ache. Oracle says that only a single thread should execute commands on a context, so I am not sure if OCIBreak is an exception to that rule. I may just try it and see what happens, but just because it works does not mean it is right or even stable. I'd like definitive proof (or disproof), so if anyone has some insight that'd be welcome.
Thanks again.
Edited by: DreamWarrior1337 on Jan 22, 2009 2:53 PM