The invokeLater loop

[Stefan Ram]

I invented what I call the »invokeLater loop«:

class Object implements java.lang.Runnable
{ int i = 0;
public void run()
{ java.lang.System.out.println( i++ );
javax.swing.SwingUtilities.invokeLater( this ); }}

public class Main
{ public static void main( final java.lang.String[] args )
{ javax.swing.SwingUtilities.invokeLater( new Object() ); }}

What is it good for?

Maybe it provides a way to loop while still allowing
Swing events to be processed in parallel? (So you do not
always have to spawn another thread besides the EDT.)

One also might see it as a kind of recursion that will not
overflow that stack. But since it does not atually nest
incarnations, it is not really adopted for recursive programming.

 

[Daniel Pitts]

I invented what I call the »invokeLater loop«:

class Object implements java.lang.Runnable
{ int i = 0;
public void run()
{ java.lang.System.out.println( i++ );
javax.swing.SwingUtilities.invokeLater( this ); }}

public class Main
{ public static void main( final java.lang.String[] args )
{ javax.swing.SwingUtilities.invokeLater( new Object() ); }}

What is it good for?

Maybe it provides a way to loop while still allowing
Swing events to be processed in parallel? (So you do not
always have to spawn another thread besides the EDT.)

One also might see it as a kind of recursion that will not
overflow that stack. But since it does not atually nest
incarnations, it is not really adopted for recursive programming.

A few comments:

That's isn't really a new invention.

It isn't recursive.

It encourages coupling the application logic to the UI environment.

It is a good way to eat up CPU cycles if you don't want to have your CPU
go idle. Also known as a laptop warmer.

It doesn't take advantage of concurrent processing capabilities of most
modern systems, ultimately slowing down both your EDT and your
application processing.

It is more error prone and harder to tune than using a SwingWorker.

Unrelated to the point you made Stefan, I feel I should make a note on
Stefan Ram(TM) convention. When I see "new Object()", and it isn't
referring to java.lang.Object, it takes me 10 times as long to
understand what the code is doing. I literally stared at that for 3 or 4
seconds before it clicked that you were using confusing class names and
hoping your abstruse convention would "clarify" the confusion.

 

[Stefan Ram]

It is a good way to eat up CPU cycles if you don't want to have your CPU
go idle. Also known as a laptop warmer.

Say, I am writing an implementation of the »Game Of Life«. I have an
endless main loop whose contents updates to the next generation.

I want to have a sequence like:

- calculate next generation
- update screen
- calculate next generation
- update screeen
- calculate next generation
- ...

I think that I can do this by calculating the next generation and
updating the screen in »paintComponent« and then »invokingLater«
paintComponent or requesting a screen update which indirectly will
call »paintComponent« again. This will give the sequence above.

Another possibility might be to calculate the next generation in
a separate thread (SwingWorker). But I am not sure how to get
the above sequence (without waiting an arbitrary period of
times or complicated handshaking across thread boundaries).

 

[Daniel Pitts]

Say, I am writing an implementation of the »Game Of Life«. I have an
endless main loop whose contents updates to the next generation.

I want to have a sequence like:

- calculate next generation
- update screen
- calculate next generation
- update screeen
- calculate next generation
- ...

I think that I can do this by calculating the next generation and
updating the screen in »paintComponent« and then »invokingLater«
paintComponent or requesting a screen update which indirectly will
call »paintComponent« again. This will give the sequence above.

Depending on the type of simulation, you might not actually want to wait
for the screen to update before continuing. This is actually
particularly true in multi-player games. GoL may be an exception of
course, since typically users do want to see every frame.

By the way, calling "repaint()" doesn't mean the screen is updated.
You'd have to be actively updating the screen each loop.

Another possibility might be to calculate the next generation in
a separate thread (SwingWorker). But I am not sure how to get
the above sequence (without waiting an arbitrary period of
times or complicated handshaking across thread boundaries).

Calling SwingWorker.invokeAndWait() isn't that complicated of a boundary
IMO.

public class GoLMain {
public static void main(String...args) {
GameOfLifeSimulation gols = new GameOfLifeSimulation();
GolUI ui = initUi();
while (true) {
Generation g = gols.nextGeneration();
SwingUtilities.invokeAndWait(ui.generationPainterFor(g));
}
}
}

 

[Joerg Meier]

Unrelated to the point you made Stefan, I feel I should make a note on
Stefan Ram(TM) convention. When I see "new Object()", and it isn't
referring to java.lang.Object, it takes me 10 times as long to
understand what the code is doing. I literally stared at that for 3 or 4
seconds before it clicked that you were using confusing class names and
hoping your abstruse convention would "clarify" the confusion.

By this point, that must be intentional obfuscation, outright designed and
intended to cause bugs and present a maintenance nightmare, and you won't
convince me otherwise. Overwriting "Object" is so purely malicious that
there is no way that it's just random ineptitude, especially not from
Stefan, who certainly isn't clueless enough to not be aware of that.

My best guess would be that it's a straw man in the hopes of someone
complaining about a bug it caused, which could then be responded to with
something along the lines of "If only you would fully qualify every class
reference like me, this wouldn't have happened to you, this proves that
imports are evil".

But since Stefan generally does not address criticism of his coding style,
we will most likely forever be left to wonder what particular insanity
sparks the Stefan Ram(TM) convention

Liebe Gruesse,
Joerg

 

[Eric Sosman]

Say, I am writing an implementation of the »Game Of Life«. I have an
endless main loop whose contents updates to the next generation.

I want to have a sequence like:

- calculate next generation
- update screen
- calculate next generation
- update screeen
- calculate next generation
- ...

I think that I can do this by calculating the next generation and
updating the screen in »paintComponent« and then »invokingLater«
paintComponent or requesting a screen update which indirectly will
call »paintComponent« again. This will give the sequence above.

How long does "calculate the next generation" take?

- If it takes "significant" time, you're tying up the EDT
and making the GUI unresponsive during the calculation.

- If it takes "insignificant" time, you might as well just
go ahead and do the deed without all the fuss and bother.

Either way, I see no use case for the scheme. (Also, I don't
think you can rely on having one paintComponent() call per update:
If your GUI gets minimized everything might just freeze, or if
another window obscures part of your GUI you might get multiple
paintComponent() calls to refresh the display, or multiple calls
with different clip regions to paint around the intruder. You
need some extra work to make sure paintComponent() only calls
invokeLater() "when appropriate.")

Another possibility might be to calculate the next generation in
a separate thread (SwingWorker). But I am not sure how to get
the above sequence (without waiting an arbitrary period of
times or complicated handshaking across thread boundaries).

Leapfrogging a Runnable from EDT moment to EDT moment with
multiple invokeLater() self-invocations ... What a splendid way
to avoid complication!

It'd be simple enough to do it with a SwingWorker, if you
wanted to. The GUI would create a SwingWorker instance to compute
one generation and call execute() to start it, running its
doInBackground() method on a worker thread. When doInBackground()
finishes, the SwingWorker's done() method runs on the EDT, where
it can paint the display, and create and launch another SwingWorker
for the next generation. Easy-peasy.

Personally, I'd do it by driving things from the calculation
thread: Compute a new generation, compute the changes, send the
changes to the EDT via invokeLater() while the calculation thread
starts work on the next generation. Easy to control the frames-
per-second rate, too, which I think would be horribly difficult
using EDT-multiplexing.

Perhaps there's an application for the leapfrog technique, but
I've not yet heard or thought of one.

 

[Daniele Futtorovic]

Say, I am writing an implementation of the »Game Of Life«. I have an
endless main loop whose contents updates to the next generation.

I want to have a sequence like:

- calculate next generation
- update screen
- calculate next generation
- update screeen
- calculate next generation
- ...

I think that I can do this by calculating the next generation and
updating the screen in »paintComponent« and then »invokingLater«
paintComponent or requesting a screen update which indirectly will
call »paintComponent« again. This will give the sequence above.

Not sure I get this. If you are performing non-UI-related computations
in your "Object" (I second Daniel's remarks on your choice of
nomenclature, btw), then these will happen on the EDT. But they should
actually be happening /off/ the EDT.

So it seems to me either you are busying the EDT and lagging the app, or
it's really just a laptop warmer. Perhaps another example would be useful.

 

[markspace]

Maybe it provides a way to loop while still allowing
Swing events to be processed in parallel? (So you do not
always have to spawn another thread besides the EDT.)

Swing events can't be processed in parallel, they're run on a single thread.

 

[Silvio]

I invented what I call the »invokeLater loop«:

class Object implements java.lang.Runnable
{ int i = 0;
public void run()
{ java.lang.System.out.println( i++ );
javax.swing.SwingUtilities.invokeLater( this ); }}

public class Main
{ public static void main( final java.lang.String[] args )
{ javax.swing.SwingUtilities.invokeLater( new Object() ); }}

What is it good for?

Maybe it provides a way to loop while still allowing
Swing events to be processed in parallel? (So you do not
always have to spawn another thread besides the EDT.)

One also might see it as a kind of recursion that will not
overflow that stack. But since it does not atually nest
incarnations, it is not really adopted for recursive programming.

What is with this horrible code layout. Really makes makes me not want
to read the code.

 

[Daniel Pitts]

[snip]
What is with this horrible code layout. Really makes makes me not want
to read the code.

Don't bother, Stefan is a long time poster refusing to cater to this
audience.

It is better than some, however. Not quite "troll level". Try reading
some of qwertymonkey's code for instance.

 

[Roedy Green]

Maybe it provides a way to loop while still allowing
Swing events to be processed in parallel? (So you do not
always have to spawn another thread besides the EDT.)

You want Swing manipulation to run on the EDT and non Swing stuff to
run on some other threads. invokeLater lets you put stuff on the
Swing thread from some non-Swing thread. If you are already on the
Swing thread, you don't need it.

 

[Stefan Ram]

I wrote on 2013-05-10:

I invented what I call the »invokeLater loop«:
class Object implements java.lang.Runnable
{ int i = 0;
public void run()
{ java.lang.System.out.println( i++ );
javax.swing.SwingUtilities.invokeLater( this ); }}
public class Main
{ public static void main( final java.lang.String[] args )
{ javax.swing.SwingUtilities.invokeLater( new Object() ); }}
What is it good for?

I the meantime, I have actually used the invokeLater loop in
an animation. From the functional requirements point of view,
it did work. But it was by orders of magnitude slower than
the same animation with a for loop. OTOH, the for loop blocks
the EDT (if you do not want to start another thread, that is).
So, I ended up using the for loop as an inner loop to do
some iterations fast and then using invokeLater to call the
next iterations of the for loop, so that the EDT gets some
leeway to do other things. So I got my animation with a
single thread of execution, running fast and the EDT still
being responsive.

 

[Knute Johnson]

I wrote on 2013-05-10:

I invented what I call the »invokeLater loop«:
class Object implements java.lang.Runnable
{ int i = 0;
public void run()
{ java.lang.System.out.println( i++ );
javax.swing.SwingUtilities.invokeLater( this ); }}
public class Main
{ public static void main( final java.lang.String[] args )
{ javax.swing.SwingUtilities.invokeLater( new Object() ); }}
What is it good for?

I the meantime, I have actually used the invokeLater loop in
an animation. From the functional requirements point of view,
it did work. But it was by orders of magnitude slower than
the same animation with a for loop. OTOH, the for loop blocks
the EDT (if you do not want to start another thread, that is).
So, I ended up using the for loop as an inner loop to do
some iterations fast and then using invokeLater to call the
next iterations of the for loop, so that the EDT gets some
leeway to do other things. So I got my animation with a
single thread of execution, running fast and the EDT still
being responsive.

That doesn't make a lot of sense to me. Calling invokeLater schedules a
task on the EDT. If you want something to run consistently, calling
invokeAndWait would make more sense. But given modern computers have
multiple processors with multiple cores, it really makes more sense to
use a separate thread and do active rendering off the EDT or to use
invokeAndWait and keep your rendering times short. Both will allow time
for the EDT to process other things and give you fast and smooth animation.