Pausing Quantum Renderer
kevin.rushforth at oracle.com
Thu Nov 19 21:20:30 UTC 2015
This might be a tricky semantic to achieve, and great care will be
needed to ensure no deadlocks or race conditions. Not running any more
pulses after this method returns seems fine, but it might be better to
let any existing renderJobs run (possibly discarding the results). This
way you could send the pause message to the renderer as a special
renderJob and not have to worry about jobs that are scheduled but not
Johan Vos wrote:
> After some experiments, here is my current thinking:
> Toolkit can have 2 new methods:
> When pauseRenderer is called, it should be guaranteed that after this call,
> no new pulses are fired until resumeRenderer is called.
> That is not hard, but it is not enough. Before we pause the pulses, the
> previous pulse probably submitted a renderJob to Prism, executed on the
> QuantumRenderer ThreadPoolExecutor. That job should run fine, as the next
> pulse (when we're back) will call GlassScene.waitForRenderingToComplete().
> So we have to wait until there are no running or pending tasks in the
> QuantumRenderer as well.
> - Johan
> On Wed, Nov 18, 2015 at 9:58 PM, David Hill <David.Hill at oracle.com> wrote:
>> On 11/18/15, 3:45 PM, Johan Vos wrote:
>> I think that it would be reasonable to put in something to Quantum
>> that causes the render loop to "pause", and then resume later. I envision
>> this toggle as causing the render loop to skip, rather than tinkering with
>> the pulses.
>> When resume is called, it might be best to treat the world as dirty.
>> Added to Toolkit, this would allow someone like Monocle to make the
>> toggles as is appropriate.
>> If this works for you, perhaps you could prototype it ?
>>> On Android, a JavaFX Application might transfer control to another app
>>> become invisible) and enter the foreground later. In that case, the
>>> GLSurface we are rendering on becomes invalid. In order to avoid problems
>>> and save battery, we want to pause the renderer thread, but this turns out
>>> to be more difficult than I expected.
>>> When our app transfers control, we get a callback from Android. We
>>> intercept this in javafxports, and we set the Screen width/height to 0/0
>>> we don't want to render on the (invalid) surface while we lost control.
>>> When we regain control, we resize the Screen and the app renders again.
>>> The reason we set the width/height to 0/0 is because the PresentingPainter
>>> will call SceneState.isValid() and this returns false in case getWidth()
>>> getHeight() are 0.
>>> However, SceneState extends PresentableState and it overrides the update
>>> method. It will call PresentatbleState.update() which will set the
>>> viewWidth to the width of the new Screen (hence, 0) , but after that it
>>> overwrites the viewWidth with camera.getViewWidth(). The latter still
>>> contains the old value. A quick inspection shows that
>>> is called when validate(...) is called on NGDefaultCamera, which is called
>>> by ES2Context.updateRenderTarget() which happens during rendering, hence
>>> *after* the PresentingPainter checks if the width is 0.
>>> So immediately after we set the width of the Screen to 0 (on the FX App
>>> Thread), a Pulse happens, and this one still things the screen is the
>>> original size. While the pulse is happening, the android system destroys
>>> our context, and the rendering fails. Moreover, the EGL system is in a
>>> unpredictable state (recreating the surface fails).
>>> A very dirty workaround for this is to wait for 1 pulse (with the new
>>> pulselistener API this should be possible) before we return from the
>>> callback method called by Android when the surface is about to be
>>> destroyed. That way, we will have 1 bogus rendering on an existing (but
>>> about-to-be-destroyed) surface.
>>> But it would be better if there is some way to tell the quantum renderer
>>> immediately stop rendering. Existing pulses are no problem, as the
>>> renderLock guarantuees that we set the size to 0 only when no other thread
>>> (quantum renderer) has a lock on the renderLock.
>>> - Johan
>> David Hill<David.Hill at Oracle.com>
>> Java Embedded Development
>> "A man's feet should be planted in his country, but his eyes should survey
>> the world."
>> -- George Santayana (1863 - 1952)
More information about the openjfx-dev