Drastically reduce the number of created strings in d8, 2nd attempt.

Drastically reduce the number of created strings in d8, 2nd attempt.

This also reduces the number of TLS accesses, so it is a double-win. We have
230k less TLS accesses while running Octane now, so we are down to 23% now
compared to the start of these TLS-related CLs.

To get things right in the presence of multiple Isolates, we have to thread the
correct Isolate through several layers. This threading wasn't that bad after
all, it keeps one honest about the real depdencies. The only ugly thing is that
in ExternalArrayWeakCallback we have to conjure up the current Isolate from the
TLS, but this is a know API deficiency.

Introduced a tiny helper function for throwing on the way.

Committed: https://code.google.com/p/v8/source/detail?r=13018

[Michael Starzinger, 2012-11-20 14:35:46]

LGTM with a nit.

https://codereview.chromium.org/11418088/diff/4001/src/d8.cc
File src/d8.cc (right):

https://codereview.chromium.org/11418088/diff/4001/src/d8.cc#newcode107
src/d8.cc:107: #undef DISPOSE_SYMBOL
Can we defensively reset the embedder data to NULL here?

[Sven Panne, 2012-11-20 14:46:22]

Landing...

https://codereview.chromium.org/11418088/diff/4001/src/d8.cc
File src/d8.cc (right):

https://codereview.chromium.org/11418088/diff/4001/src/d8.cc#newcode107
src/d8.cc:107: #undef DISPOSE_SYMBOL
On 2012/11/20 14:35:46, Michael Starzinger wrote:
> Can we defensively reset the embedder data to NULL here?

Done.