Introduce a PrototypeIterator template and use it all over the place

Introduce a PrototypeIterator template and use it all over the place

The idea is to have one central place where all prototype access happens.
The different template parameters specialize on

- How to objects are stored (raw pointer vs handle)
- Whether we walk the prototype chain based on the map, or via a type switch
- Where the prototype chain ends (null value, non hidden prototype,
  given object)

A macro is provided for the common case that only the first prototype is
accessed.

BUG=???
R=dcarney@chromium.org,verwaest@chromium.org
LOG=n

[jochen (gone - plz use gerrit), 2014-06-24 09:53:19]

ptal

I haven't converted cctests and map->prototype() yet

[Toon Verwaest, 2014-06-25 12:44:40]

Can't you derive STORE_AS_HANDLE / STORE_AS_POINTER from the input type?

I find the names TYPE_BASED_WALK / MAP_BASED_WALK quite confusing. They are just
artifacts from our current implementation right?
What I think you want is to just have 2 versions: one which takes JSObjects as
input, and one which takes Objects as input. From Object we first derive the
"root" as I did in lookup.[h|cc], and then walk the prototype JSObject-based
chain starting from that root. So we only need to use GetPrototype(Isolate*)
once for translating Object* to JSObject.

The result of GetCurrent() probably should be a JSReceiver since it could be a
proxy. When current is JSProxy, we are at end as well.

The only actual manual configuration we need, I think, is where to stop.

[jochen (gone - plz use gerrit), 2014-06-25 13:41:48]

On 2014/06/25 at 12:44:40, verwaest wrote:
> Can't you derive STORE_AS_HANDLE / STORE_AS_POINTER from the input type?

I'd need two fields in the class instead of one.

All methods (IsAtEnd/Advance) have different implementations, so I'd always need
to check which field is used.

I fear the compilers won't optimize this away in all cases.

> 
> I find the names TYPE_BASED_WALK / MAP_BASED_WALK quite confusing. They are
just artifacts from our current implementation right?
> What I think you want is to just have 2 versions: one which takes JSObjects as
input, and one which takes Objects as input. From Object we first derive the
"root" as I did in lookup.[h|cc], and then walk the prototype JSObject-based
chain starting from that root. So we only need to use GetPrototype(Isolate*)
once for translating Object* to JSObject.
> 
> The result of GetCurrent() probably should be a JSReceiver since it could be a
proxy. When current is JSProxy, we are at end as well.
> 
> The only actual manual configuration we need, I think, is where to stop.