[racket-dev] gui responsiveness

Thanks for the great (and quick!) response.  It's good to know that the 
ordering is intentional, and to have some nice ways to work around it if 
needed.

The reason I thought that refreshes were lower priority was because of 
the scrollbar behavior in the program below.  On Unix/X, dragging the 
scrollbar back and forth does lots of paints, but I only actually see a 
few of them.

Does that sound like a bug to you?


#lang racket/gui

(define num-on-paint 0)

(define frame
   (new frame%
        (label "Refresh")
        (width 500)
        (height 500)))

(define (draw-screen canvas dc)
   (set! num-on-paint (add1 num-on-paint))
   (sleep 0.1) ; simulate a longer painting effort
   (send dc draw-text (~a num-on-paint " paints") 400 70))

(define canvas
   (new canvas%
        (parent frame)
        (paint-callback draw-screen)
        (style '(no-autoclear hscroll))))

(send frame show #t)

(send canvas init-auto-scrollbars 700 #f 0.0 0.0)



On 04/16/2014 01:56 PM, Matthew Flatt wrote:
> You're right that it's about event ordering and not refresh coalescing.
> Since mouse events are handled after refreshes, you won't get the next
> refresh request until an earlier one is handled, after which the next
> mouse event can trigger another refresh request. I think the difference
> between Unix/X and Windows may be that Windows sends fewer mouse
> events.
>
> There are trade-offs here, but my experience is that ordering input
> events before refresh does not work well in general. To trigger
> refreshes at a lower priority in this case, you could use Neil's
> suggestion or change
>
>   (send this refresh)
>
> to
>
>   (set! needed? #t)
>   (queue-callback (lambda () (when needed?
>                                (set! needed? #f)
>                                (send this refresh)))
>                   #f) ; => low priority
>
> where `needed?` is a field that's initially #f.
>
> At Wed, 16 Apr 2014 13:33:02 -0400, David Vanderson wrote:
>> (moved to dev)
>>
>> On Linux, the attached program shows terrible responsiveness when
>> dragging points around on the graph.  Can anyone else on Linux reproduce
>> this behavior?
>>
>>
>> The patch below dramatically improves the responsiveness by forcing the
>> eventspace to process medium-level events (mouse movement) before
>> refresh events.  Without the patch, each mouse drag causes a paint.
>> With it, multiple mouse drags are processed before a paint.
>>
>>
>> I'm unsure about this fix.  Windows doesn't show the problem (I don't
>> have a mac to test), so I think it's just a GTK issue.
>>
>> My guess is that the gui layer is relying on the native libraries to
>> coalesce multiple refresh requests (but this is not working with GTK).
>> Can anyone confirm this?
>>
>> Thanks,
>> Dave
>>
>>
>>
>> diff -ru
>> racket-6.0_clean/share/pkgs/gui-lib/mred/private/wx/common/queue.rkt
>> racket-6.0/share/pkgs/gui-lib/mred/private/wx/common/queue.rkt
>> --- racket-6.0_clean/share/pkgs/gui-lib/mred/private/wx/common/queue.rkt
>> 2014-02-18 12:27:43.000000000 -0500
>> +++ racket-6.0/share/pkgs/gui-lib/mred/private/wx/common/queue.rkt
>> 2014-04-16 09:41:16.810993955 -0400
>> @@ -300,8 +300,8 @@
>> (lambda (_) #f))
>>                                               (or (first hi peek?)
>> (timer-first-ready timer peek?)
>> -                                               (first refresh peek?)
>>                                                   (first med peek?)
>> +                                               (first refresh peek?)
>>                                                   (and (not peek?)
>>                                                        sync?
>>                                                        ;; before going
>> with low-priority events,
>>
>>
>> ------------------------------------------------------------------------------
>> [text/plain "graph_ui.rkt"] [~/Desktop & open] [~/Temp & open]
>> _________________________
>>    Racket Developers list:
>>    http://lists.racket-lang.org/dev