Concurrency issue in foreach tasks

[kikofernandez]

There seems to be a concurrency / parallelism issue when running the test async_foreach and async_force_gc. The consequence of this issue is a deadlock of the program.

[albertnetymk]

Comparing the cycle detector with the task runner, I can't find where task runner is set as system actor. Maybe this is the reason, partially? If so, it's good opportunity to extract task runner out like how cycle detector is implemented.

[kikofernandez]

the task runner is not a system actor. Currently, the task runner has the same type as the Main active actor. The reason for this was to re-use the main actor's dispatcher method

[kikofernandez]

@albertnetymk could you confirm that you get an assert error when popping an actor from the scheduler? I have tested this on Mac and Linux and that is not the case running vagrant.

In my case, there seems to be a data race that sets the number of scheduler_waiting to a value bigger than scheduler_count. This breaks the quiescent algorithm and the program ends up in an infinite loop.

Working on it, but would like to confirm that there's some consistency between platforms.

[albertnetymk]

Sometimes this shows:
async_force_gc: sched/scheduler.c:330: run: Assertionpop(sched) == ((void *)0)' failed.`

I used --ponythreads 16, btw.

[kikofernandez]

ok, thanks! I have never seen than one yet, but I'll keep it in mind.