Improve performance using array instead of linked-list to implement Backbone.Events

[lifesinger]

In Chrome 18:

the linked-list version: 97,270  57% slower
the array version: 227,369 fastest

http://jsperf.com/backbone-events-linked-list/8

Maybe it is worthy to merge this pull request.

[jashkenas]

It was originally changed from an array implementation to a linked list implementation for performance reasons. Mind taking a look at the original ticket, and letting me know what you think?

[lifesinger]

@jashkenas Thanks for your information, I found this ticket:

#707

And the source code of the array implementation before the linked list version is:

https://github.com/braddunbar/backbone/blob/9982e314cbacb94429a838e480be20bb111cb73e/backbone.js

This array implementation has some "ugly" code to affect performance, such as:

1. place slice.call in a loop
2. use [[callback, context], [callback, context], ...] data structure
3. leave null hole when unbind

In my array implementation, I have deal with carefully to fix all the performance issues.

A updated performance comparision:

http://jsperf.com/backbone-events-linked-list/8
http://jsperf.com/backbone-events-linked-list/9

It seems that the new array-implementation is very faster than the current linked-list version.

And more important, the array implementation is more readability than the linked-list version. I think it is worthy coming back to the array implementation with performance improving.

[braddunbar]

This is fantastic stuff @lifesinger! Sorry about the original request. I shouldn't have been so quick to close it. :)

I think the huge discrepancies in Chrome and Firefox are due to their copy-on-write array implementations. In the vast majority of cases the callback list won't be altered during trigger so this is a huge win. I'm going to merge this and make a couple of stylistic changes. Wonderful stuff, thanks @lifesinger!

[lifesinger]

@braddunbar It is very happy to see this merge.

Except copy-on-write array implementations in chrome and firefox, another notable performance improving is:

changing

rest = slice.call(arguments, 1);

to

for (i = 1, length = arguments.length; i < length; i++) {
        rest[i - 1] = arguments[i];
      }

In most cases, the arguments.length is less than 5. So, use loop to set rest is more effective than slice.call. For readability, I suggest to add a comment before this loop.

Hope that Backbone is getting better and better.

[braddunbar]

Agreed. I was rather surprised that slice.call(arguments, 1) is so slow. It was accounting for almost half the time spent in trigger for small callback lists. Thanks for this @lifesinger!