WIP: new.target

WIP: new.target

BUG=v8:3887

[arv (Not doing code reviews), 2015-02-24 22:22:55]

arv@chromium.org changed reviewers:
+ dslomov@chromium.org

[arv (Not doing code reviews), 2015-02-24 22:22:55]

Very basic support. It only works for derived constructors at the moment.
However, the parsing stuff is all done.

I'm not really sure how to split this up or how to proceed from here. Maybe
implementing the whole semantics at once is a better approach?

[Dmitry Lomov (no reviews), 2015-02-25 11:28:41]

This factoring looks ok to me. 
Implement for derived constructors first and then proceed from there.

https://codereview.chromium.org/949763003/diff/1/src/full-codegen.cc
File src/full-codegen.cc (right):

https://codereview.chromium.org/949763003/diff/1/src/full-codegen.cc#newcode846
src/full-codegen.cc:846: // to execute the constructor and only fail when an
assignment to 'this'
I believe we can now kill this entire function and all its uses (since now the
parser guarantees that 'super(...)' only occurs where new.target is avaliable.

https://codereview.chromium.org/949763003/diff/1/src/scopes.cc
File src/scopes.cc (right):

https://codereview.chromium.org/949763003/diff/1/src/scopes.cc#newcode315
src/scopes.cc:315: new_target_ =
I'd prefer not declaring new.target if it is not needed.

[arv (Not doing code reviews), 2015-02-25 14:47:25]

https://codereview.chromium.org/949763003/diff/1/src/full-codegen.cc
File src/full-codegen.cc (right):

https://codereview.chromium.org/949763003/diff/1/src/full-codegen.cc#newcode846
src/full-codegen.cc:846: // to execute the constructor and only fail when an
assignment to 'this'
On 2015/02/25 11:28:40, Dmitry Lomov (chromium) wrote:
> I believe we can now kill this entire function and all its uses (since now the
> parser guarantees that 'super(...)' only occurs where new.target is avaliable.

That is unrelated to this CL. I'll remove it in another CL.

https://codereview.chromium.org/949763003/diff/1/src/scopes.cc
File src/scopes.cc (right):

https://codereview.chromium.org/949763003/diff/1/src/scopes.cc#newcode315
src/scopes.cc:315: new_target_ =
On 2015/02/25 11:28:40, Dmitry Lomov (chromium) wrote:
> I'd prefer not declaring new.target if it is not needed.

This is where I was when I got pulled into other work. My plan was to do this
all conditionally but it fails because scope_uses_new_target_ was not getting
applied correctly. At one point I thought I could do the parsing part in one CL.
However, the parser needs new_target_ to be allocated (theres a DCHECK iirc).

This is where the CL is incomplete. The tracking of new.target is not working
correctly.

https://codereview.chromium.org/949763003/diff/1/src/scopes.cc#newcode318
src/scopes.cc:318: new_target_->AllocateTo(Variable::PARAMETER, -2);
Do you think this is OK for cases like?

function f() {
  print(new.target);
}
new f()

[Dmitry Lomov (no reviews), 2015-02-26 14:15:05]

https://codereview.chromium.org/949763003/diff/1/src/scopes.cc
File src/scopes.cc (right):

https://codereview.chromium.org/949763003/diff/1/src/scopes.cc#newcode318
src/scopes.cc:318: new_target_->AllocateTo(Variable::PARAMETER, -2);
On 2015/02/25 14:47:24, arv wrote:
> Do you think this is OK for cases like?
> 
> function f() {
>   print(new.target);
> }
> new f()

Depends on the codegen we will have for such functions.