Variance implementation: Inference fails on tests

[eernstg]

variance_in_inference_test

The test language/variance/variance_in_inference_test fails with the CFE and the analyzer, in several locations.

Inference does not produce the right outcome in at least some of these failure locations:

tests/language/variance/variance_in_inference_test.dart:52:66: Error: The argument type 'Contravariant<Middle>' can't be assigned to the parameter type 'Contravariant<Upper>'.
 - 'Contravariant' is from 'tests/language/variance/variance_in_inference_test.dart'.
 - 'Middle' is from 'tests/language/variance/variance_in_inference_test.dart'.
 - 'Upper' is from 'tests/language/variance/variance_in_inference_test.dart'.
  var inferredMiddle = inferContraContra(Contravariant<Upper>(), Contravariant<Middle>());
                                                                 ^

- Line 52, column 66: COMPILE_TIME_ERROR.ARGUMENT_TYPE_NOT_ASSIGNABLE
  The argument type 'Contravariant<Middle>' can't be assigned to the parameter type 'Contravariant<Upper>'.

The inferred expression is inferContraContra(Contravariant<Upper>(), Contravariant<Middle>()), which allows for T == Middle, but the inference actually chooses Upper, and then there is an error because a Contravariant<Middle> can't be assigned to the resulting parameter type Contravariant<Upper>. The constraints are T <: Upper and T <: Middle, so inference should find the solution T == Middle.

variance_inout_inference_test

Similarly, the test language/variance/variance_inout_inference_test gives rise to an error based on failing inference:

tests/language/variance/variance_inout_inference_test.dart:46:40: Error: The argument type 'Invariant<Middle>' can't be assigned to the parameter type 'Invariant<Upper>'.
 - 'Invariant' is from 'tests/language/variance/variance_inout_inference_test.dart'.
 - 'Middle' is from 'tests/language/variance/variance_inout_inference_test.dart'.
 - 'Upper' is from 'tests/language/variance/variance_inout_inference_test.dart'.
  var inferredMiddle4 = inferInvContra(Invariant<Middle>(), Contravariant<Upper>());
                                       ^

- Line 46, column 40: COMPILE_TIME_ERROR.ARGUMENT_TYPE_NOT_ASSIGNABLE
  The argument type 'Invariant<Middle>' can't be assigned to the parameter type 'Invariant<Upper>'.

variance_in_inference_error_test

A third case with the same issue is language/variance/variance_in_inference_error_test:

variance_in_inference_error_test.dart:32:66: Error: The argument type 'Contravariant<Middle>' can't be assigned to the parameter type 'Contravariant<Upper>'.
 - 'Contravariant' is from 'variance_in_inference_error_test.dart'.
 - 'Middle' is from 'variance_in_inference_error_test.dart'.
 - 'Upper' is from 'variance_in_inference_error_test.dart'.
  var inferredMiddle = inferContraContra(Contravariant<Upper>(), Contravariant<Middle>());
                                                                 ^

error • The argument type 'Contravariant<Middle>' can't be assigned to the parameter type 'Contravariant<Upper>'. • variance_in_inference_error_test.dart:32:66 • argument_type_not_assignable

Again, inference should produce the solution T == Middle from the constraints T <: Upper and T <: Middle, but it chooses T == Upper and then complains. The same issue arises in several other locations in this test.

variance_downwards_inference_test

Finally, the test language/variance/variance_downwards_inference_test reports three spurious errors:

variance_downwards_inference_test.dart:42:25: Error: The argument type 'List<num>' can't be assigned to the parameter type 'List<int>'.
 - 'List' is from 'dart:core'.
  B<int> b = new B(<num>[])..x=2.2;
                        ^
variance_downwards_inference_test.dart:42:32: Error: A value of type 'double' can't be assigned to a variable of type 'int'.
  B<int> b = new B(<num>[])..x=2.2;
                               ^
variance_downwards_inference_test.dart:52:38: Error: A value of type 'double' can't be assigned to a variable of type 'int'.
  D<int> d = new D(3, (num x) {})..x=2.2;
                                     ^

  error • Couldn't infer type parameter 'T'.

Tried to infer 'int' for 'T' which doesn't work:
  Parameter 'x' declared as     'List<T>'
                but argument is 'List<num>'.
The type 'int' was inferred from:
  Return type declared as 'B<T>'
              used where  'B<int>' is required.

Consider passing explicit type argument(s) to the generic.

 • variance_downwards_inference_test.dart:42:18 • could_not_infer
  error • The argument type 'List<num>' can't be assigned to the parameter type 'List<int>'. • variance_downwards_inference_test.dart:42:20 • argument_type_not_assignable
  error • A value of type 'double' can't be assigned to a variable of type 'int'. • variance_downwards_inference_test.dart:42:32 • invalid_assignment
  error • A value of type 'double' can't be assigned to a variable of type 'int'. • variance_downwards_inference_test.dart:52:38 • invalid_assignment

In line 42, the CFE should choose T == num, but it chooses T == int, and the two errors are then implied. T == int is not a solution to the expected constraints, so it seems likely that the generated constraints are not correct. The analyzer chooses int as well, and then detects that it causes errors, and still reports those errors even though they are derived from a choice made during type inference.

In line 52, they should choose T == num, but they actually choose T == int, which causes the reported error. T == int is a solution, but the heuristic for contravariant parameters should choose the most general solution num.

[eernstg]

Refresh as of April 2023: https://dart-ci.firebaseapp.com/current_results/#/filter=language/variance shows that almost every configuration is failing with several tests, including:

• language/variance/variance_downwards_inference_test
• language/variance/variance_in_inference_test
• language/variance/variance_inout_inference_test

[kallentu]

Fixed recently with 166f83d.
Thanks for filing this issue Erik!