Killing threads with exception executes finally clauses

[jopereira]

Original issue 14 created by jopereira on 2013-08-01T13:58:20.000Z:

Since SimulationThreadDeath is a Throwable/Error any finally statements in real code that lie on its path up the stack will be executed upon closing the Process/Host. This of course leads to unwanted behavior.

(Reported by N. Stifter)

[jopereira]

Comment #1 originally posted by jopereira on 2013-08-01T14:27:08.000Z:

There are three alternative implementations:

• Kill threads with our own SimulationThreadDeath exception.
• Kill threads with Thread.stop(), leading to ThreadDeath exception.
• Don't kill threads, just quiesce them.

The problem with not killing threads is that we leak resources. In cases where a large number of simulation worlds is used (e.g., to run a large number of unit tests), this is not acceptable.

The other alternatives will invoke finally clauses. Even if one could re-write bytecode to avoid finally clauses in instrumented classes, we would still have stack frames from original classes that we can't avoid. This solution is thus incomplete.

The better solution should be to completely sandbox each of the host instances, such that whatever happens when the stack is unwound, does not produce external effects. Namely, we should be able to discard all output.