dbstop if error stops in "drawnow"

Dear Community,

I'm facing a rather odd problem and want to know if someone experienced a similar problem, or even knows a solution.

I usually have

dbstop if error

enabled. In a script that prints png-files with bargraphs the debugger stops in

36  drawnow; % give changes a chance to be processed

The Function Call Stack is: Base * myscript * print alternatePrintPath

Quit Debugging or Continue often results in another stop after 5-10 seconds of processing.

My script does nothing fancy, the only "unusual" thing in my mind might be that the struct from which the pngs are plotted is over 1 GB in size. The break by the debugger is not labeled as an error just as shown above as "36 drawnow ..." in the internal print.m

Without "dbstop if error" the script just runs fine.

[N/A]

Adding a

pause(0.02);  % Avoid debugger stops in Matlab 2017a.
              % Walter has suggested it in the forum.
              % Does not help after PRINT, but before
              % ?*? Review 2018

before the print prevents the debugger from stopping. Walter suggested it in a comment below.

This circumvents the problem, be careful with such magic :)

Answer 1

[Jan]

drawnow stops with an error, if a graphic object has been created, which cannot be displayed correctly, e.g. a listbox with the 'Value' set higher than the number of elements. Or a slider with the value outside the defined range. Or A listbox with a single selection activated, but the value is empty.

Then updating the graphics fail "inside" drawnow. But you should get a corresponding error message directly. But you do not mention any message, but only the calling stack.

What does "results in another stop after 5-10 seconds" exactly mean? Where does it stop?

What does "The break by the debugger is not labeled as an error" mean and what is the "internal print.m"?

You wrote that you "usually" run your code with the enabled dbstop. Are you aware that Matlab's JIT accelerator is disabled then? Some loops might be up to 100 times faster without enabled debugger.

Can you post a piece of code, which reproduces the error? Currently the problem looks really strange and magic, but this means usually, that something trivial happens that the programmer is not aware of. Without seeing the code, it is hard to guess such details.

[N/A]

I do not get a red error message, just this line appears:

36  drawnow; % give changes a chance to be processed

It stops apparently in the same place and will look like:

>> plot_TTL_ID(struct2use,[savePath])
36  drawnow; % give changes a chance to be processed
36  drawnow; % give changes a chance to be processed
36  drawnow; % give changes a chance to be processed
36  drawnow; % give changes a chance to be processed
K>>

The "plot_TTL_ID" contains this piece of code

figure
bar(uniqueCount(:,1)) % bargraph
xticks(1:length(uniqueCount(:,1))) % show all ticks
xticklabels(uniqueLabel) % label all ticks
xtickangle(90) % turn label 45° for better readability
XaX = get(gca,'XTickLabel');
set(gca,'XTickLabel',XaX,'FontName','Times','fontsize',6)

print(gcf,'-dpng','-r300',savePath);

This often but not always stops in "drawnow", without the debugger the script does not stop.

With "internal print.m" I mean that the script is from the graphics toolbox of matlab2017a and with the break by debugger I mean that the debugger stops as if an error was encountered but none is displayed.

[Walter Roberson]

Try putting a drawnow() before the print(). If that does not work, try putting a pause(.2) before the print()

[Jan]

Just a note: size(uniqueCount, 1) is faster than length(uniqueCount(:,1)), which creates the temporary column vector at first.

Why do you get the the TickLabels from gca only to set it again in the next line?

The pause(0.02) helps to let Java perform some internal stuff. It is a mystic command, because nobody could explain, why is works. pause(0.01) does not work. Use such magic tricks with great care, because it is a bad programming practice. Write a meaningful comment:

pause(0.02);  % Avoid debugger stops in Matlab 2017a.
              % Walter has suggested it in the forum.
              % Does not help after PRINT, but before
              % ?*? Review 2018

I add a key also, to remember to review the line later, if a new Matlab version still needs it. With every new Matlab version I search for all "?*?" keys in the code and check if they are still needed.

This converts a bad programming practice in a clear documented bad programming practice. :-) And if it works, it is fine.

"Does not help after PRINT, but before" was a guess only! I'm not sure where the problem is coming from.

By the way: Comments are an important and valuable part of the code. They can reduce the debug time by hours and allow for maintaining the code. Keep them clean. "xtickangle(90)" does not rotate by 45°.

[N/A]

Thank you for your addition,

The pause magic works.

My matlab knowledge is mainly self taught thus I appreciate such comments!

I should use "size" more often. I set the TickLabels again because I copied that piece from another code and haven't thought about it, I apologize. The 45° is not updated, it was 45° once.

I added the % comment as suggested, the ?*? is a neat idea I'll use it from now on whenever I use some other magic ;)

Matlab runs on a server and I connect via a NoMachineClient, could that be the reason why the pause works?

[Jan]

The pause(0.02) seems to trigger the processing of the Java event queue. The drawings of objects is not performed, when they are created, but only when a screen update is triggered. This reduces the number of expensive updates e.g. for:

H = plot(1:10, rand(1,10));
set(H, 'YData', -rand(1-10));

If the screen is updated directly after the first command, the double work is needed. Therefore Matlab waits until a pause or drawnow command occurs, or the CommandWindow is activated.

Many (or all?) graphic objects are implemented in Java, which has an own event queue. Sometimes the processing of this queue is not finished, when Matlab expects it to be, e.g. if you use getframe on a figure, which contains uicontrols. Sometimes the uicontrols like buttons are not drawn yet and getframe records the pixels from the underlying window or desktop background instead. Then a pause(0.02) helps also.

Magic.