DOMTokenList: Reuse index from DOMTokenList instance if possible …

[amakhrov]

…instead of parsing raw string attribute every time

In our Angular Universal app we extensively use utility classes that can be toggled conditionally.
While investigation recent performance issues, I noticed that the single longest-running piece of code is getList function from DOMTokenList. Looks like adding multiple classes to a dom element one by one is O(N^2), as at each step getList is called and parses the raw string property into a list of tokens. This parsing seems to be quite a heavy operation.

This PR implements a way to avoid unnecessary parsing by reusing results from the previous step.

In our project this change reduced the class list processing time by half.

[alan-agius4]

Nice 😊

[alan-agius4]

@cscott, can you please take a look at this? This PR is really important for the Angular team.
Thanks.

[cscott]

This doesn't actually fix the O(N^2) behavior. You'd need to keep the hash of elements from getList() around so that the uniqueness check in add is O(1) in order to avoid O(N^2) behavior. But really you probably don't want the # of elements in the class string to be very large in the first place, so I'm guessing the O(N^2) aspect isn't what you're actually concerned with, but rather the constant factor. This patch looks fine for reducing the constant factor.

[amakhrov]

Good observation @cscott ! This is still N^2 with a much smaller constant.
The reason I chose not to cache the parsed array (which would be the ultimate solution) is to be more memory-efficient. For some consumers memory might be more important than cpu, and I just didn't want to make this decision :)