co19/LanguageFeatures/Patterns/type_inference_A26_t01

[scheglov]

FAILED: dart2analyzer-none release_arm64 co19/LanguageFeatures/Patterns/type_inference_A26_t01
Expected: Pass
Actual: CompileTimeError

--- Command "dart2analyzer" (took 21ms):
DART_CONFIGURATION=ReleaseARM64 sdk/bin/dart --packages=.dart_tool/package_config.json pkg/analyzer_cli/bin/analyzer.dart --use-analysis-driver-memory-byte-store --dart-sdk=/Users/scheglov/Source/Dart/sdk.git/sdk/sdk -Dtest_runner.configuration=custom-configuration-1 --enable-experiment=patterns,records --ignore-unrecognized-flags --packages=/Users/scheglov/Source/Dart/sdk.git/sdk/.dart_tool/package_config.json --format=json /Users/scheglov/Source/Dart/sdk.git/sdk/tests/co19/src/LanguageFeatures/Patterns/type_inference_A26_t01.dart

unexpected analysis errors in type_inference_A26_t01.dart:
- Line 31, column 10: COMPILE_TIME_ERROR.LIST_ELEMENT_TYPE_NOT_ASSIGNABLE
  The element type 'double' can't be assigned to the list type 'int'.

- Line 32, column 22: COMPILE_TIME_ERROR.TYPE_ARGUMENT_NOT_MATCHING_BOUNDS
  'num Function(num)' doesn't conform to the bound 'int Function(int)' of the type parameter 'R'.

- Line 44, column 26: COMPILE_TIME_ERROR.MAP_VALUE_TYPE_NOT_ASSIGNABLE
  The element type 'double' can't be assigned to the map value type 'int'.

- Line 45, column 22: COMPILE_TIME_ERROR.TYPE_ARGUMENT_NOT_MATCHING_BOUNDS
  'num Function(num)' doesn't conform to the bound 'int Function(int)' of the type parameter 'R'.

main() {
  num a = 0;
  a as int;
  a.isEven;
  [a] = [3.14];

My understanding is that we are expected to use the promoted type as the context type, so [3.14] becomes <int>[3.14].

As I can see, the specification says "The context type schema for a pattern ... Identifier: In an assignment context, the context type schema is the static type of the variable that p resolves to."

It does says later about type checking, but this is after type schema was used: "In an assignment context, the required type of p is the (unpromoted) static type of the variable that p resolves to." This makes [a] = <double>[3.14] valid.

[sgrekhov]

Hm... I think that here we should use unpromoted type of a which is num (and therefore no error should occur). @eernstg please confirm

[eernstg]

My understanding is also that [a] = [3.14] should give rise to an error at 3.14, because the list literal on the right will infer as <int>[3.14].

I believe that <num>[a] = <num>[3.14] should not have any compile-time errors.

We should be able to use <double>[a] = <double>[3.14], too: It is allowed to assign to a of type num by having it as an identifier pattern in a pattern assignment (but <String>[a] = ... is an error because String is the matched value type for a, and String isn't assignable to the required type num of the identifier pattern).

<num>[a] = <double>[3.14] should also work.

Various combinations which will become one of the above after type inference should also work, e.g., <num>[a] = [3.14].

[sgrekhov]

Thank you. But there is no error in the case of record. Is this expected?

  var (num b,) = (0,);
  b as int;
  b.isEven;
  (b,) = (3.14,); // no error

[eernstg]

That's OK: The context type schema of (b,) is (num). This makes no difference for the initializing expression (3.14,), but it gets static type (double), and that's assignable to the required type (Object?) of the pattern. Then we need to type check b with matched value type double, and there's no error because double is assignable to num.

So the difference between the list pattern and the record pattern is basically that we aren't deciding on a list of type arguments determining the types of fields of the record (because records don't accept type arguments at all, they are more like an infinite number of pre-defined classes with any given field types, and with subtyping relationships whenever we have the same shape plus a subtype relationship for every field). In contrast, we are using the list type argument to get to <int>[3.14], which is subsequently an error.