[Isolates] Avoid isolate lookup in HandleScope destructor.

[Isolates] Avoid isolate lookup in HandleScope destructor.

This has a nice side effect that we can assert the isolate stayed the
same. I had to update one of our new tests to satisfy this assert.

Committed: http://code.google.com/p/v8/source/detail?r=5578

[Dmitry Titov, 2010-09-29 22:13:59]

LGTM!

I wonder if we should have something more drastic then ASSERT, since in case the
HandleScope API is not used correctly, as was in the test before, in release
build it'll try to use deallocated Isolate and not always fail... Maybe
API_Fatal or make HandleScope::Leave a noop if the underlying Isolate was
released?

[Dmitry Titov, 2010-09-29 22:14:47]

Also (still LGTM):

http://codereview.chromium.org/3545004/diff/1/2
File include/v8.h (right):

http://codereview.chromium.org/3545004/diff/1/2#newcode474
include/v8.h:474: inline void Initialize(internal::Isolate* i) {
Tiny nit, to consider, but I'm 50-50 on it myself:

It feels that Initialize is used more like "Reset" (at least in api.cc).

If the Data had a constructor taking Isolate*, then we could remove call to
Initialize from isolate.cc and don't do change in api.cc.

[Vitaly Repeshko, 2010-10-01 17:16:49]

http://codereview.chromium.org/3545004/diff/1/2
File include/v8.h (right):

http://codereview.chromium.org/3545004/diff/1/2#newcode474
include/v8.h:474: inline void Initialize(internal::Isolate* i) {
On 2010/09/29 22:14:47, Dmitry Titov wrote:
> Tiny nit, to consider, but I'm 50-50 on it myself:
> 
> It feels that Initialize is used more like "Reset" (at least in api.cc).
> 
> If the Data had a constructor taking Isolate*, then we could remove call to
> Initialize from isolate.cc and don't do change in api.cc.

When Data is used in an API HandleScope it can't be constructed with the right
arguments and stays uninitialized for a while. So it's more natural to
initialize it than to reset.

[Dmitry Titov, 2010-10-01 17:59:51]

You are right. Never mind :-)