Yes it works well, but it's not really necessary. It's inefficient wasted effort.
I guess it's fun to make lovely APIs, but that doesn't help people who just want their js to run fast.
This doesn’t happen. The object is unchanged when you call array.slice.
> Maybe runtimes optimize
Only insofar as the built-in array methods are optimized anyway.
> maybe they create completely new objects
No, they don’t. Or rather, yes, array.slice creates a new object to put the slice in, but only because that is part of its declared API.
> Yes it works well, but it's not really necessary.
What’s the alternative?
> It's inefficient wasted effort.
Inefficient how? If you time it, it’s faster other ways of accomplishing the same thing. Whose wasted effort? It certainly saves me time as a JavaScript developer.
* * *
To be honest, I think that arguments should just be an array (then we wouldn’t have any of this trouble), but I assume that’s not possible for the language at this point.
You're wrong. Why don't you check instead of reading API docs or saying what you think should happen? It obviously is changed, and is a new object: Array.prototype.slice.call({0:"hello",1:"world",foo:"bar",length:2});
Results in ["hello","world"] - the result is a completely new object - inefficient wasted effort. Both in cpu and memory.
Offtopic again, but here's an interesting way to make a new N element array, initialized with undefineds...
Array.prototype.slice.call({length:100}) = [undefined, undefined, undefined........ undefined]
Array.prototype.slice.call({length:10000000}) takes almost 1 second to run on my MBP while it's creating that new array.
>> What’s the alternative?
As I say, just write closures...
onclick = function(c){return function(){c.doSomething()}}(this);
It's not much more typing, and is more general. I agree arguments would be nicer if it was a true array though.
Also, I can live with defining such an argument on initial definition, as with Python’s self in method definitions, but having to add an extra argument to every function/method invocation for the execution scope, that must be passed in every time explicitly or everything breaks, is really obnoxious. You could certainly make up a convention for the proper name, but just using this is more straight-forward, and doesn’t really have any particularly significant downsides.
There’s a reason that pretty much every JavaScript framework includes a bind function/method, and that they’re adding it to the official language spec: it’s a convenient and effective pattern.