Hello everyone,
I'd like to discuss a not-very-easy to understand topic, like interrupt checking mechanism in squeak VM.
I introduced an event system in HydraVM: all external events to Interpreter can be posted at any time using event queue without much processor stall (i'm using atomic operations for that). I really now considering to change interpret() loop to move out and cleanup code concerning interrupts checking.
How it's currently done (how do i see it - please tell me if i'm wrong with this): there is an interruptCheckCounter which initialized to big number, and in different places of interpret(), there are method, called quickCheckForInterrupts, which decrements the counter, and if it falls beyond zero, calls the checkForInterrupts() method which doing all gory details by checking many things , like user input events and any external events like signaling semaphores. In the end, it resets interruptCheckCounter back to big number. There's also a functions like forceInterruptCheck, which simply sets counter to zero, making checkForInterrupts be called as soon as possible.
How i'm suggesting this can be done: - change a quickCheckForInterrupts() to check if interpreter's event queue is not empty, and if so, return from interpret() , and outside there will be a code which will handle all pending events, and then enter interpret() again.
- for periodical interrupt checking we could use a system timer, which simply post an events periodically, making calls of checkForInterrupts() at same time intervals. This is more better than using an interruptCheckCounter, it will do checks with more precise/regular intervals.
- for any other external events all you need to do is to post a new event to interpreter's event queue. Again, it's better because your event will be handled as soon as possible, AND!! without calling common checkForInterrupts() procedure, which is a bit cumbersome, in cases if your only intent was to signal a semaphore.