Implement the pushContext & popContext methods

[Hywan]

Fix #16.

Tests are missing.

[Hywan]

Documentation is missing too.

[staabm]

initial pushContext would only make sense, if a popContext would also be triggered "automatically". but this seems to be not possible, so I would assume the caller should call both explicitly so the callers' code is more logical.

wdyt?

[Hywan]

We can call the popContext in the destructor, but when the object is destructed, the SplStack will be destructed just before, so no need to pop the first (so the latest) context :-). Nop?

[staabm]

popContext in the descructor would not be a good design (because it might never be called, e.g. when the object is assigned e.g. to a property). Therefore I think the __construct should not have side-effects like this. The class should define in its contract that pushContext and popContext need to be called etc.

[Hywan]

Actually, the first push is to set elementMap because its default value is null. So where is the problem?

[staabm]

hm think you are right.. didn't read carefully enough to see that there is a "init" case for that on pushContext

[evert]

I'd like to take a stab at this myself a bit later in the day =) there's a few other things I would like to change up a bit...

[Hywan]

@evert As you like :-). I have added inheritance and the stack includes elementMap and baseUri as requested.

[evert]

@Hywan

I looked into SplStack. It's kind of cool, but overall I think using it is a little bit less legible. So one reason to go for it, is that it perhaps has better performance features.

So I wrote a benchmark : array vs SplStack. Doing a few pushes and pops.

End result

    Method Name   Iterations    Average Time      Ops/second
    ---  ------------  --------------    -------------
    spl: [100,000   ] [0.0000080101895] [124,840.99107]
    arr: [100,000   ] [0.0000042340684] [236,179.46311]

One thing tha'ts immediately obvious, is that the operations are so fast, that it's almost silly to optimize for. However, using a plain array is twice as fast so that means I have no argument for SplStack anymore aside from that it's kinda fun to use specialized spl objects.

I didn't test for memory, but I doubt SplStack would come out well there.

[Hywan]

Good :-).