Make String::Empty inlineable.

Make String::Empty inlineable.

R=svenpanne@chromium.org
TEST=cctest/test-api/StringEmpty

Committed: https://code.google.com/p/v8/source/detail?r=11429

[Sven Panne, 2012-04-25 06:56:21]

LGTM. Just a little general remark: With every single change like this we say a
little bit more "goodbye" to linking v8 dynamically, because we lose more and
more binary compatibility. But I guess this was never a design goal...

https://chromiumcodereview.appspot.com/10199019/diff/1/src/api.cc
File src/api.cc (right):

https://chromiumcodereview.appspot.com/10199019/diff/1/src/api.cc#newcode4633
src/api.cc:4633: if (!EnsureInitializedForIsolate(isolate,
"v8::String::Empty()")) {
Just out of curiosity: Why do we need this change? Or asked another way: Why do
we *not* need it in tons of other places? Or should we test the result of
EnusreInitializedForIsolate everywhere? api.cc is extremely inconsistent...

[Michael Starzinger, 2012-04-25 08:18:31]

About dynamic linking: You are right, I can imagine that shipping V8 as a shared
library is a "hard job" for distributions at the moment.

The first thing we would need to make this easier, is to introduce an API
version number scheme. Whenever we land a change in v8.h, those version number
need/might to be bumped accordingly.

But my change doesn't make the situation worse, it just introduces more reasons
why the aforementioned version numbers would need to be bumped. And we have
plenty of reasons in v8.h already.

https://chromiumcodereview.appspot.com/10199019/diff/1/src/api.cc
File src/api.cc (right):

https://chromiumcodereview.appspot.com/10199019/diff/1/src/api.cc#newcode4633
src/api.cc:4633: if (!EnsureInitializedForIsolate(isolate,
"v8::String::Empty()")) {
On 2012/04/25 06:56:22, Sven Panne wrote:
> Just out of curiosity: Why do we need this change? Or asked another way: Why
do
> we *not* need it in tons of other places? Or should we test the result of
> EnusreInitializedForIsolate everywhere? api.cc is extremely inconsistent...

Yes, IMHO we need this change in tons of other places. Because
EnsureInitializedForIsolate() can return even though the Isolate is not
initialized or is already dead, in cases where the embedder installed a fatal
error handler with SetFatalErrorHandler() that doesn't terminate the
application.

This just wasn't tested for String::Empty() before. I'll try to fix this
incrementally with my changes. We could also fix this for all functions in one
sweep, I you think that's worth the trouble.