Fix inference bugs found by sra@ by changing how we merge if/then/else locals.

Fix inference bugs found by sra@ by changing how we merge if/then/else locals.

R=sra@google.com

Committed: https://code.google.com/p/dart/source/detail?r=26171

[sra1, 2013-08-14 21:35:19]

lgtm

https://chromiumcodereview.appspot.com/22955003/diff/1/sdk/lib/_internal/comp...
File sdk/lib/_internal/compiler/implementation/types/inferrer_visitor.dart
(right):

https://chromiumcodereview.appspot.com/22955003/diff/1/sdk/lib/_internal/comp...
sdk/lib/_internal/compiler/implementation/types/inferrer_visitor.dart:394:
fieldScope.mergeDiamondFlow(thenBranch.fieldScope, elseBranch.fieldScope);
Is there a case where the fieldScope should be merged when elseBranch == null?

I could not get a bug from:

var y = x && (field = 1);

but I don't understand why this code is correct.

[ngeoffray, 2013-08-15 07:01:34]

Thanks Stephen.

https://chromiumcodereview.appspot.com/22955003/diff/1/sdk/lib/_internal/comp...
File sdk/lib/_internal/compiler/implementation/types/inferrer_visitor.dart
(right):

https://chromiumcodereview.appspot.com/22955003/diff/1/sdk/lib/_internal/comp...
sdk/lib/_internal/compiler/implementation/types/inferrer_visitor.dart:394:
fieldScope.mergeDiamondFlow(thenBranch.fieldScope, elseBranch.fieldScope);
On 2013/08/14 21:35:19, sra1 wrote:
> Is there a case where the fieldScope should be merged when elseBranch == null?

No, you only need to merge if the two branches are there. If there is only one
branch, initializations of fields happening in the branch don't matter because
they are under a condition. If there are two branches, both branches must
initialize the field to ensure the field will not be null.

> 
> I could not get a bug from:
> 
> var y = x && (field = 1);
> 
> but I don't understand why this code is correct.

The control flow looks like this:

   [if x]
    | \
    |  \
    | [field = 1]
    |   /
    |  /
   [....]

So whatever happens to 'field' in the branch doe snot ensure it won't be null.

[ngeoffray, 2013-08-15 08:04:36]

Committed patchset #1 manually as r26171 (presubmit successful).