>> That's fixable. There are a couple of places in SystemWindow's reframe 
>> code where it starts using "Sensor cursorPoint" and "Sensor 
>> anyButtonPressed" (partly in the Rectangle code that does the fast 
>> framing).
>
> Hmm, that's not so pretty, but the methods should still work, 
> shouldn't they?

They _do_ work, who claims that they don't and on what grounds?

> Maybe my caveat from the last message is a good idea in the current
> design, too: have EventSensor simulate the polling primitives by
> remembering the most recent event.  That should be easier on 
> the VM than using a combination of the primitives.

Which is exactly what EventSensor does.

Cheers,
  - Andreas

PS. Before starting to make well-meant suggestions it's a good idea to check
out what's actually going on in the system. For example, a Very Worthwhile
Exercise (tm) is to print out the value of "Sensor" - it has been an
EventSensor for longer than you might be aware of. Any code relying on
Sensor in any form has been working for years (including event synthesis
from polling and non-polling apis into both directions) so any newly
introduced oddities are quite unlikely to originate from there.