Destroy the runners cleanly when shutting down the engine.

Adds a new engine_step_prop_stop which makes the runners quit if set. engine_clean now sets the prop, releases the runners, and pthread_joins them.

Fixes #559 (closed).

changed the description

Just one thought. Wouldn't it be safer to wait for all the runners to exit before proceeding to clean up the caches, i.e. put the caches in a second loop?

Not really... The caches are runner-specific, so nobody else should be able to access them except for the runner itself. At least that's my understanding of how they are used.

Yes, so aren't we, in principle anyway, removing something a runner could be using before we know for certain it has exited (since k can be less than the runner in question). Not going to happen if all threads are at the barrier, but the sanitizer broke that contract somehow.

Exactly, the pthread_join above blocks until the runner's thread has exited :)

Thanks, forgot that was a blocking call. Makes sense now.

OK, all looks to be working and the sanitizer is back to reporting embarrassing leaks, rather than embarrassing crashes. Accepting.

merged

mentioned in commit 535917ea