Partially revert r2761....

Partially revert r2761.

Do not create handles for values of internal fields---this operation is performance
critical and plain pointers are safe.

Appy the same approach to External wrapping and unwrapping.

Plus some minor refactorings.


Committed: http://code.google.com/p/v8/source/detail?r=3064

[Christian Plesner Hansen, 2009-10-13 14:24:17]

See comments

[Christian Plesner Hansen, 2009-10-13 14:28:08]

Here are the real comments

http://codereview.chromium.org/270085/diff/1001/2002
File src/api.cc (right):

http://codereview.chromium.org/270085/diff/1001/2002#newcode2582
Line 2582: if (as_object->IsSmi()) {
Doesn't this give a different result from External::Wrap?  Using different
encodings seems really dangerous to me.

[antonm, 2009-10-13 14:30:37]

Christian,

thanks a lot for comments.

http://codereview.chromium.org/270085/diff/1001/2002
File src/api.cc (right):

http://codereview.chromium.org/270085/diff/1001/2002#newcode2582
Line 2582: if (as_object->IsSmi()) {
On 2009/10/13 14:28:08, Christian Plesner Hansen wrote:
> Doesn't this give a different result from External::Wrap?  Using different
> encodings seems really dangerous to me.

It does indeed.  And it indeed dangerous, however it has a notable performance
impact.  Actual, when it was discussed idea was to drop everything except for
storing native pointers into internal fields :)

[Christian Plesner Hansen, 2009-10-14 11:26:37]

http://codereview.chromium.org/270085/diff/1001/2002
File src/api.cc (right):

http://codereview.chromium.org/270085/diff/1001/2002#newcode2582
Line 2582: if (as_object->IsSmi()) {
I don't have anything against using this approach as such but then we should
change all code that encodes pointers to using this scheme.

[antonm, 2009-10-14 12:22:18]

http://codereview.chromium.org/270085/diff/1001/2002
File src/api.cc (right):

http://codereview.chromium.org/270085/diff/1001/2002#newcode2582
Line 2582: if (as_object->IsSmi()) {
On 2009/10/14 11:26:37, Christian Plesner Hansen wrote:
> I don't have anything against using this approach as such but then we should
> change all code that encodes pointers to using this scheme.

Christian, may you point out other methods?

[antonm, 2009-10-14 14:11:35]

Christian,

may you have another look?

[Christian Plesner Hansen, 2009-10-14 14:22:41]

Lgtm.