How can you identify what's triggering round of change detection detection in Angular2 (debugging ngDoCheck infinite loop)?

Question

I haven't been able to make a minimal reproduction I can put in a plunkr, but in my app I have an infinite loop where ngDoCheck in the root component is called infinitely, so I'd like to be able to identify what is actually changing.

I've looked through every call in the stack trace from putting a breakpoint in ngDoCheck but can't find anything useful.

I know that triggering change detection from within change detection can cause this kind of infinite loop (though I thought debug mode, which I have enabled, would catch it) but I can't find any instances of this manually.

Is there anything one can log or inspect that will give insight into what's causing a round of change detection?

Probably not very useful, but there's a snapshot of the stack and state of things during this loop:

I will take a look if you create reproducible plunker, otherwise it's hard to tell since there is not enough details — Max Koretskyi, Mar 31 '17 at 20:27
I suspect it is just change detection is running after some async action. Probable you have global listener or setInterval — yurzui, Mar 31 '17 at 20:27
@Maximus I guess the main part of the question is "is there a way to find out what's causing change detection?" which is independent of any reproducible plunkr - but I'm not sure if it's possible. — tobek, Mar 31 '17 at 20:37
@yurzui This change detection loop is running 100s of times per second (completely freezing UI) - I don't have any `setInterval`s running that fast nor can I imagine any async activity I'm using doing that much stuff, unless some library I'm using is going haywire... either way it would be great to just find out what's causing any given round of change detection. — tobek, Mar 31 '17 at 20:39
Put breakpoint in ZoneDelegate.invoke and observe what kind of tasks is executing. And would be great if you reproduced it in plunker or in github — yurzui, Mar 31 '17 at 20:52
@yurzui thanks for your help, but didn't find much there... here's a paste of stringifying `arguments` from breakpoint inside `ZoneDelegate.prototype.invoke` (argument names are `targetZone, callback, applyThis, applyArgs, source`): https://pastebin.com/MhZZZZSG. No info really except that it's the Angular zone (via `long-stack-trace`). — tobek, Mar 31 '17 at 22:56

score 18 · Accepted Answer · edited Jul 18 '22 at 02:28

May be a bit late, but still may be helpful for someone.

I found a couple of solution which can help you to find what actually cause a CD cycle.

First one, as was mention in comments is to check the source field of Task, the easiest way is

 // put this in any component which is rendered on the page
  public ngDoCheck() {
    console.log('doCheck', Zone.currentTask.source);
  }

This will output something like this:

This is already something and you have the direction of further researching, but it's still not perfect.

To find a real place where action handler is defined we need zone's long-stack-trace spec.

// add this, for example, in your polyfills file 
import 'zone.js/dist/long-stack-trace-zone';

Now additionally to the source of task we can check data prepared by long-stack-trace

  public ngDoCheck() {
    console.log('doCheck', Zone.currentTask.source, Zone.currentTask.data.__creationTrace__);
  }

This will output an array:

We are interested in error property of each entry in the array. Start investigating in console from 0 element, and then go further.

Inside this property will be a stack trace like this:

Maybe this way is overcomplicated, but this is what I've found by reading source code of Zone. Will be really glad if someone suggests an easier solution.

Update 1

I created a small snippet, which can help you to extract long stack traces

function renderLongStackTrace(): string {
  const frames = (Zone.currentTask.data as any).__creationTrace__;
  const NEWLINE = '\n';

  // edit this array if you want to ignore or unignore something
  const FILTER_REGEXP: RegExp[] = [
    /checkAndUpdateView/,
    /callViewAction/,
    /execEmbeddedViewsAction/,
    /execComponentViewsAction/,
    /callWithDebugContext/,
    /debugCheckDirectivesFn/,
    /Zone/,
    /checkAndUpdateNode/,
    /debugCheckAndUpdateNode/,
    /onScheduleTask/,
    /onInvoke/,
    /updateDirectives/,
    /@angular/,
    /Observable\._trySubscribe/,
    /Observable.subscribe/,
    /SafeSubscriber/,
    /Subscriber.js.Subscriber/,
    /checkAndUpdateDirectiveInline/,
    /drainMicroTaskQueue/,
    /getStacktraceWithUncaughtError/,
    /LongStackTrace/,
    /Observable._zoneSubscribe/,
  ];

  if (!frames) {
    return 'no frames';
  }

  const filterFrames = (stack: string) => {
    return stack
      .split(NEWLINE)
      .filter((frame) => !FILTER_REGEXP.some((reg) => reg.test(frame)))
      .join(NEWLINE);
  };

  return frames
    .filter(frame => frame.error.stack)
    .map((frame) => filterFrames(frame.error.stack))
    .join(NEWLINE);
}

Usage:

  public ngDoCheck() {
    console.log(renderLongStackTrace());
  }

Update 2

I found that for EventTask, zone's long-stack-trace create wrong stack traces, you can track this issue here https://github.com/angular/zone.js/issues/1195

Update 3

It's worth to set Error.stackTraceLimit = Infinity; before application bootstrap (not for production), because stack traces with angular + zone can be really looong.

How can you identify what's triggering round of change detection detection in Angular2 (debugging ngDoCheck infinite loop)?

1 Answers1

Update 1

Update 2

Update 3