Slow ng-repeat because of AnimateRunner even without ngAnimate

[ArekSredzki]

I'm trying to print a lot of elements to the DOM, but it is painfully slow.
However, the reason is caused by AnimateRunner and nothing else.

Note: I do not want to filter the elements further, but rather get to the bottom of why an un-needed component is stalling the entire browser

I have disabled ngAnimate, the page has no animations, yet the slowest part of the whole process is it's complete function.

Please see the profile below

Your suggestions are greatly appreciated.

[Narretz]

Which version of Angular is this?

[Narretz]

Why does this function exist?
the core ng module has its own implementation of the $animate service. So even when ngAnimate isn't included the functions that are called are the same: $animate.enter etc, but of course the amount of work that is done is much less. When ngAnimate is included, it overwrites these functions and adds the actual animation logic.
The runner is needed because even when ngAnimate isn't included we need to make sure that animation event listeners are executed correctly. So a certain subset of logic (the animate queue and the runner) exists in the ng module, too.
There's probably potential to make these faster, but that is the status quo. You are very welcome to see if there are improvments to be made.
That said. ngRepeat was never lightning fast and I assume with many elements (how many do you try to render?) it will always be slow.

[Narretz]

The profile also doesn't really show that the slowness is "caused by AnimateRunner and nothing else."

[frfancha]

Why animation event listeners must be run when ngAnimate is not included?

[Narretz]

The event listeners / callbacks should fire even when no actual animations run. This can be important for example if you are testing your application without animations. Then you would still want the resilts of the callbacks that fire when the animation is complete.

[ArekSredzki]

@Narretz
Thank you for the prompt response.

This is with version 1.5, running in an electron browser window (chromium).

I will complete more meaningful testing once I get into the office later today.

[frfancha]

@Narretz Thanks for the explanation.
However I still don't understand (sorry for my slowness ;-) ).
If the goal is only to enable testing animations without including ngAnimate, then why is it always (and not on demand) enabled and why should we always pay the performance penalty (including in production) when we don't intend to use ngAnimate at all?

[gkalpak]

@Narretz, do we really need the RAFs without ngAnimate included ?

[ArekSredzki]

The large majority of the computation time is spent checking whether the element should have animations applied to it. Unfortunately this occurs even if we don't have any animations... I'll try to make some improvements.

[Narretz]

The large majority of the computation time is spent checking whether the element should have animations applied to it. Unfortunately this occurs even if we don't have any animations... I'll try to make some improvements.

Can you elaborate? Which code does that? Please be specific if you mean the code with the ngAnimate module as a dependency or without it.

@gkalpak I'm not sure if the rAF is needed, but it definitely enforces asyncronicity, which is possibly what we want for consistency.

@frfancha Different example: you are using a 3rd party module that uses $animate and also an animation callback that is important. If you are now disabling ngAnimate from your app you would expect the DOM operations and the callback to work as before.

[ArekSredzki]

@Narretz After some further investigation I've identified the issue with a high count of redundant calls to retrieve the window.document object.

It is especially noticeable when using angular in electron because electron will make a quick ipc call to it's main thread to retrieve whether the document is hidden. This call is quick, but synchronous as it's integrated to getting the window.document object, and slows everything to a halt in volume.

One might argue that this is an issue with electron, but I would disagree.
In the very large majority of cases the document is not hidden and this code is a blocker in many of angular's operations (ex. ngSwitchWatchAction, $watchCollectionAction, ngIfWatchAction). Plus, it's likely that other frameworks make use of similar get implementations.

TL;DR it is unnecessary to make calls for document.window thousands of times in the span of a second for the rare case when the document is hidden.

Stats
Using current AnimateRunner._tick: 2510ms
_tick which just calls rafTick(fn): 650ms

Interested to hear your thoughts on how the redundant calls should be avoided.