game simulator stuck keyboard bug

[tballmsft]

Sometimes when I am using the game simulator with the keyboard (specifically, the arrow keys) and doing a lot of quick keystrokes (say for moving a character), one (or sometimes two) of the keys will start repeating even though I am not pressing the keys. This happens non-deterministically and basically destroys game play. The only recourse is to restart the game simulator.

[tballmsft]

On both the Arcade simulator and hardware, I code the following

• Press and hold the left-arrow key (KEY_DOWN event) for while, causing sprite to start moving left (via handler subscribed to DOWN)
• As I am holding down the left-arrow key, the game controller logic starts generating KEY_REPEAT events as well, which keep the sprite moving left (another handler)
• Release the left-arrow key (KEY_UP event), after which (according to the controller logic), KEY_REPEAT events should stop

However, sometimes the REPEAT events keep coming, even though I have release the left-arrow key.

I have traced this problem down to an inversion of handlers. In both the browser and Arcade devices, the act of holding down the left-button continuously generates KEY_DOWN events. What I am seeing in the trace though is the following

1. Long trace of KEY_DOWN events and KEY_REPEAT generated by controller
2. last KEY_DOWN event
3. last KEY_UP event
4. KEY_UP_handler invoked
5. KEY_DOWN handler invoked

The important part is the inversion of handler invocation that happens (this is the race condition): the KEY_UP handler is invoked before the KEY_DOWN handler even though the KEY_UP event occurred after the KEY_DOWN event – the result is that controller logic keeps generating KEY_REPEAT events, as the last event received was KEY_DOWN.

The usual and correct trace is:

1. Long trace of KEY_DOWN events and KEY_REPEAT generated by controller
2. last KEY_DOWN event
3. KEY_DOWN handler invoked
4. KEY_UP_handler invoked
5. last KEY_UP event

Each handler is execute by a different fiber. Given sequential code that spawns two threads, P and Q, in sequence, it certainly seems like it should be possible for the code of Q to start executing before that of P. Is the same thing possible with fibers?