Don't flatten every time we call CharCodeAt

Don't flatten every time we call CharCodeAt

Add String->GetMayFlatten(int) that tries to get a character and may flatten.
This uses an isolate global counter to flatten at regular intervals, instead
of doing it every time.

[Vitaly Repeshko, 2011-05-10 13:07:02]

http://codereview.chromium.org/6991007/diff/1/src/objects.cc
File src/objects.cc (right):

http://codereview.chromium.org/6991007/diff/1/src/objects.cc#newcode5214
src/objects.cc:5214: if (this->GetIsolate()->ShouldFlattenString()) {
Here "this" can already be a flat cons (i.e. with empty second). We shouldn't
update the isolate counter in this case. Maybe ShouldFlattenString() should take
a string argument.

[Lasse Reichstein, 2011-05-10 16:27:46]

http://codereview.chromium.org/6991007/diff/1/src/objects.cc
File src/objects.cc (right):

http://codereview.chromium.org/6991007/diff/1/src/objects.cc#newcode5214
src/objects.cc:5214: if (this->GetIsolate()->ShouldFlattenString()) {
Good point. We should probably not get here for a flat string.

[Erik Corry, 2011-05-11 19:31:59]

LGTM with the 'flat' naming fixed.

http://codereview.chromium.org/6991007/diff/6001/src/objects-inl.h
File src/objects-inl.h (right):

http://codereview.chromium.org/6991007/diff/6001/src/objects-inl.h#newcode183
src/objects-inl.h:183: uint32_t type =
HeapObject::cast(this)->map()->instance_type();
Assert that the flat string enum is zero.

http://codereview.chromium.org/6991007/diff/6001/src/objects-inl.h#newcode2029
src/objects-inl.h:2029: if (this->IsFlat()) {
There is some name confusion here.  Until now the nomenclature has been that a
cons string that has been flattened is a flat string (ie the second part has
length 0).  But now with the introduction of a new FlatString class it makes no
sense to ask whether a cons string is flat.

Do we need an 'almostflat' concept?  Or should we not use IsFlat here or
anywhere else.  It will not make use of the fact that we know it is a cons
string, so it will probably be wasting time querying the shape of the string
before looking at the second part.

I think we need to search the VM for the word flat and rename one of the
concepts.

[Vitaly Repeshko, 2011-05-12 15:54:53]

LGTM too modulo the same issue.

http://codereview.chromium.org/6991007/diff/6001/src/objects-inl.h
File src/objects-inl.h (right):

http://codereview.chromium.org/6991007/diff/6001/src/objects-inl.h#newcode2029
src/objects-inl.h:2029: if (this->IsFlat()) {
I agree that this is confusing. IsFlat is usually called to avoid flattening or
before getting a vector of chars. I think ConsString should have a new
IsFlattenedCons predicate and other usages of IsFlat should be replaced with
more specific predicates.

http://codereview.chromium.org/6991007/diff/6001/src/objects.cc
File src/objects.cc (right):

http://codereview.chromium.org/6991007/diff/6001/src/objects.cc#newcode5220
src/objects.cc:5220: int recurse_limit = 7;
Make this a (documented) constant in ConsString.

http://codereview.chromium.org/6991007/diff/6001/src/objects.cc#newcode5236
src/objects.cc:5236: cursor = String::cast(flatten_result->ToObjectUnchecked());
It's more straightforward to simply return 
Smi::FromInt(String::cast(flatten_result->ToObjectUnchecked())->Get(index))
here.