Improve speed of Utf8Write by always flattening the string first and

Improve speed of Utf8Write by always flattening the string first and
detecting the ASCII case.  Also rewrite Utf8Length to work on an
unflattened string.  Bug: http://code.google.com/p/v8/issues/detail?id=1665
Committed: http://code.google.com/p/v8/source/detail?r=9661

[Yang, 2011-10-17 09:40:51]

LGTM with comments.

http://codereview.chromium.org/8304021/diff/1/src/api.cc
File src/api.cc (right):

http://codereview.chromium.org/8304021/diff/1/src/api.cc#newcode3638
src/api.cc:3638: if (str->IsAsciiRepresentation()) {
IsAsciiRepresentation() is not reliable for cons and slices with regard to
external strings. For example, when a cons is constructed from two ascii
strings, and those two ascii strings are then externalized and converted to
two-byte, the instance type of the cons string still says that it's an ascii
string, which is what  IsAsciiRepresentation() returns.
Use IsAsciiRepresentationUnderneath() instead.

http://codereview.chromium.org/8304021/diff/1/src/objects.cc
File src/objects.cc (right):

http://codereview.chromium.org/8304021/diff/1/src/objects.cc#newcode5952
src/objects.cc:5952: if (input->IsAsciiRepresentation()) return total + to -
from;
Again, IsAsciiRepresentation() is not reliable.

[Erik Corry, 2011-10-17 12:39:09]

Added more tests, which also caught a bug!

http://codereview.chromium.org/8304021/diff/1/src/api.cc
File src/api.cc (right):

http://codereview.chromium.org/8304021/diff/1/src/api.cc#newcode3638
src/api.cc:3638: if (str->IsAsciiRepresentation()) {
On 2011/10/17 09:40:51, Yang wrote:
> IsAsciiRepresentation() is not reliable for cons and slices with regard to
> external strings. For example, when a cons is constructed from two ascii
> strings, and those two ascii strings are then externalized and converted to
> two-byte, the instance type of the cons string still says that it's an ascii
> string, which is what  IsAsciiRepresentation() returns.
> Use IsAsciiRepresentationUnderneath() instead.

I added a test to verify that this is OK.

http://codereview.chromium.org/8304021/diff/1/src/objects.cc
File src/objects.cc (right):

http://codereview.chromium.org/8304021/diff/1/src/objects.cc#newcode5952
src/objects.cc:5952: if (input->IsAsciiRepresentation()) return total + to -
from;
On 2011/10/17 09:40:51, Yang wrote:
> Again, IsAsciiRepresentation() is not reliable.

I think in this case it is exactly what I need.  I only need the length in UTF8,
which won't change if a string is externalized.

[piscisaureus, 2011-10-21 02:05:35]

This change is great for node.

On our http_simple benchmark with 1mb sized responses throughput jumped from
0.005 gbit/s to 0.547 gbit/s.

Btw, we create this response body by adding characters one by one. See
https://github.com/joyent/node/blob/95d530f2b4faa57662d3b052e7637246b6a7025f/...