Re: [Squeak-VMdev] primitive timing issues

[Tim R. <ti...@su...>]

I've found a relatively simple and I think plausible solution.

I factored out the forcing of an interrupt check but implemented it for now as
setting the interruptCheckCounter to -1000. In checkForInterrupts I also
factored out a test (and implemented it as 'is iCC < -100') so that forcing a
check no longer affects the feedback reset value. That at least stops
it climbing to ludicrous values. There is a small potential cost in the
check being done a bit more frequently. If we need to the force and
test can be macroized to let platforms have more choice viz Mac using a
cheap timer or whatever.

I've also, for now, changed the nextPollTick increment to 100 mS instead of
500, to give some chance of a timely response to an interruptKey press during a
heavy non-interactive process. We could quite reasonably make this settable
via vmParameter or a platform macro if that is a problem.

I made interruptChecksEveryNms vmParameter settable. I don't think the
default of 3mS is very helpful towards achieving 1mS Delay response
unless one wants to rely on enough forced interrupt checks being made
so I changed to 1mS.

Surprisingly few numbered prims need to force a check. Most that do
involve some GC so I put the forcing there. Some platform support
functions will be time consuming and the force can be added in those
places easily. For example, on RISC OS a getNextEvent that results in
polling the OS events is likely to take a while if some other app sucks
up the cycles. I have the impression from Ian that X11 clipboard
fetching can be traumatically long winded, so he can add a force to
that code. This strikes me as better than imposing a forced check on
every platform for all these things (io* and so on).

I've also altered the ioMSecs() wrap handling code in
checkForInterrupts. I think it was broken previously though given that
the timer probably wraps every six days or so it is hardly a major
problem.

As I mentioned on the general list, I'm reasonably certain that Ned's
recursive interrupt problem is not exacerbated by these changes. 

tim
--
Tim Rowledge, ti...@su..., http://sumeru.stanford.edu/tim
Useful Latin Phrases:- Recedite, plebes! Gero rem imperialem! = Stand aside plebians! I am on imperial business.