Removing watches from inside a watch can cause dirty watches to be skipped (regression)

[rino65]

Due to recent changes that added optimization in the digest function with lastDirtyWatch
if(watch === lastDirtyWatch) {....}
#5287

The scenario
during digest , one watch was called that caused a second watch to be removed. The first watch had an index which was larger then the second watch in the watchers array so the remove shifted all the indexes to the left , and the next iteration when getting length--; watchers[length] it will return the previous watch that was already called and so the condition watch === lastDirtyWatch will be true and it breaks from the loop and miss calls to the remaining watches in the array (some which have also changed)

one option to solve this is to nullify lastDirtyWatch in the unregister function that is returned from $watch

    return function() {
      arrayRemove(array, watcher);
      lastDirtyWatch = null;
    };

other is to defer the removal of the watcher till after the digest loop .

[IgorMinar]

wow.. nice report!

[IgorMinar]

resetting the lastDirtyWatch sounds like a good enough approach to me. it's not ideal as it will cause this optimization to be disabled, but if we find that these scenarios are common enough, we can create an optimization that won't break on watch removal.

[ryan65]

Hey Igor,
That's just great
thanks