Fail inference when an instance field is referenced.

Fail inference when an instance field is referenced.

This is work in progress, I will clean up and document code before
committing. But I'd like to know your opinion before moving forward.

The new top-level inference rules don't allow using instance fields
to perform type inference. So, we need to make this change.

R=brianwilkerson@google.com, paulberry@google.com
BUG=

Committed: https://github.com/dart-lang/sdk/commit/72c0c779376472a4592892ee9f9b4e85415aca94

[Brian Wilkerson, 2017-03-17 22:39:42]

lgtm

https://codereview.chromium.org/2754423002/diff/1/pkg/analyzer/test/src/summa...
File pkg/analyzer/test/src/summary/linker_test.dart (right):

https://codereview.chromium.org/2754423002/diff/1/pkg/analyzer/test/src/summa...
pkg/analyzer/test/src/summary/linker_test.dart:351: '(D) → dynamic');
Are these tests worth keeping? If we're not doing any inference then we're only
checking for standard resolution, which we have presumably tested elsewhere.

[scheglov, 2017-03-18 01:54:52]

https://codereview.chromium.org/2754423002/diff/1/pkg/analyzer/test/src/summa...
File pkg/analyzer/test/src/summary/linker_test.dart (right):

https://codereview.chromium.org/2754423002/diff/1/pkg/analyzer/test/src/summa...
pkg/analyzer/test/src/summary/linker_test.dart:351: '(D) → dynamic');
On 2017/03/17 22:39:42, Brian Wilkerson wrote:
> Are these tests worth keeping? If we're not doing any inference then we're
only
> checking for standard resolution, which we have presumably tested elsewhere.

It is partially a question of taste.
Tests in this file (linker_test.dart) are closer to unit tests.
Parsing, summarizing and linking is finely grained orchestrated by the test
itself, there is special test specific machinery for this.

Tests in the other file (top_level_inference_test.dart) are shifted closer to
functional tests.
We use AnalysisDriver to parse files, create unlinked summaries, linking and
resynthesizing.
No additional machinery in the test itself, but more happens under hood.

I incline to review these tests and move them to top_level_inference_test at
some point.



As for other tests that I touched in this CL in general, that's a fair question.
Many of them test features that are now not allowed. Unfortunately most of these
tests are bulky and test many things at once. So, careful surgery is required to
severe parts that we still want to test from the parts that are not necessary
anymore. I plan to review these tests and make these changes before landing this
CL.

[scheglov, 2017-03-18 03:46:45]

PTAL

I've done clean ups I intended to do and moved many tests.

[Brian Wilkerson, 2017-03-18 16:24:51]

lgtm

[scheglov, 2017-03-18 21:00:34]

Description was changed from

==========
Fail inference when an instance field is referenced.

This is work in progress, I will clean up and document code before
committing. But I'd like to know your opinion before moving forward.

The new top-level inference rules don't allow using instance fields
to perform type inference. So, we need to make this change.

R=brianwilkerson@google.com, paulberry@google.com
BUG=
==========

to

==========
Fail inference when an instance field is referenced.

This is work in progress, I will clean up and document code before
committing. But I'd like to know your opinion before moving forward.

The new top-level inference rules don't allow using instance fields
to perform type inference. So, we need to make this change.

R=brianwilkerson@google.com, paulberry@google.com
BUG=

Committed:
https://github.com/dart-lang/sdk/commit/72c0c779376472a4592892ee9f9b4e85415aca94
==========

[scheglov, 2017-03-18 21:00:35]

Committed patchset #2 (id:20001) manually as
72c0c779376472a4592892ee9f9b4e85415aca94 (presubmit successful).