Make closures in constructor initializers read type variables directly.

Make closures in constructor initializers read type variables directly.

BUG=dartbug.com/11540
R=johnniwinther@google.com

Committed: https://code.google.com/p/dart/source/detail?r=24674

[Johnni Winther, 2013-07-01 18:56:00]

lgtm

[karlklose, 2013-07-02 07:05:53]

PTAL, I removed the check for initializers.

[Johnni Winther, 2013-07-02 07:10:09]

SLGTM

Do we have a test where type variables are access from a closure within the
constructor body, i.e. something that exercises the non-initializer code-path?

[karlklose, 2013-07-02 07:21:40]

Thanks, Johnni!

On 2013/07/02 07:10:09, Johnni Winther wrote:
> Do we have a test where type variables are access from a closure within the
> constructor body, i.e. something that exercises the non-initializer code-path?

Yes, for example language/closure_in_constructor_test.

[karlklose, 2013-07-02 08:56:13]

Committed patchset #4 manually as r24674 (presubmit successful).

[ngeoffray, 2013-07-12 12:09:25]

DBC

https://codereview.chromium.org/18181009/diff/29001/sdk/lib/_internal/compile...
File sdk/lib/_internal/compiler/implementation/ssa/builder.dart (right):

https://codereview.chromium.org/18181009/diff/29001/sdk/lib/_internal/compile...
sdk/lib/_internal/compiler/implementation/ssa/builder.dart:3280: // not stored
on a type annotation in the closure translator. Remove when
Not sure I understand this comment. What does it mean to be stored on a type
annotation? Isn't it the resolver that does this? How come the closure
translator does something that we have to rely on here, instead of doing the
same?

[karlklose, 2013-07-12 12:22:54]

https://codereview.chromium.org/18181009/diff/29001/sdk/lib/_internal/compile...
File sdk/lib/_internal/compiler/implementation/ssa/builder.dart (right):

https://codereview.chromium.org/18181009/diff/29001/sdk/lib/_internal/compile...
sdk/lib/_internal/compiler/implementation/ssa/builder.dart:3280: // not stored
on a type annotation in the closure translator. Remove when
The wording isn't the best, I agree. With storing a type on an annotation i mean
having an entry in the types map in the tree elements. I did not mean to say
that the closure translator stores a type but that it finds an annotation for
which no types has been stored (in the resolver).

[ngeoffray, 2013-07-12 12:27:39]

https://codereview.chromium.org/18181009/diff/29001/sdk/lib/_internal/compile...
File sdk/lib/_internal/compiler/implementation/ssa/builder.dart (right):

https://codereview.chromium.org/18181009/diff/29001/sdk/lib/_internal/compile...
sdk/lib/_internal/compiler/implementation/ssa/builder.dart:3280: // not stored
on a type annotation in the closure translator. Remove when
On 2013/07/12 12:22:54, karlklose wrote:
> The wording isn't the best, I agree. With storing a type on an annotation i
mean
> having an entry in the types map in the tree elements. I did not mean to say
> that the closure translator stores a type but that it finds an annotation for
> which no types has been stored (in the resolver).

So that's a bug in the resolver?

[karlklose, 2013-07-12 12:34:16]

On 2013/07/12 12:27:39, ngeoffray wrote:
>
https://codereview.chromium.org/18181009/diff/29001/sdk/lib/_internal/compile...
> File sdk/lib/_internal/compiler/implementation/ssa/builder.dart (right):
> 
>
https://codereview.chromium.org/18181009/diff/29001/sdk/lib/_internal/compile...
> sdk/lib/_internal/compiler/implementation/ssa/builder.dart:3280: // not stored
> on a type annotation in the closure translator. Remove when
> On 2013/07/12 12:22:54, karlklose wrote:
> > The wording isn't the best, I agree. With storing a type on an annotation i
> mean
> > having an entry in the types map in the tree elements. I did not mean to say
> > that the closure translator stores a type but that it finds an annotation
for
> > which no types has been stored (in the resolver).
> 
> So that's a bug in the resolver?

More in the enqueuer, when a type annotation is resolved while its enclosing
element is not in the queue, the tree elements mapping is not stored. I debugged
this a while ago, and if I remember correctly, it was the backend triggering
resolution of signatures for helpers that are added before resolution.