Should a downwards context of dynamic be treated as empty during inference?
#32357
Comments
|
From the original issue, some examples: T foo<T>(T x) {
print(T);
return x;
}
List<T> bar<T>(T x) {
print(T);
return [x];
}
List<S> baz<S, T>(Map<S, T> x) {
print("S = $S, T = $T");
}
void main() {
var y0 = foo(4);
dynamic y1 = foo(4);
Object y2 = foo(4);
print("Bar");
var z0 = bar(4);
List<dynamic> z1 = bar(4);
List<Object> z2 = bar(4);
print("[foo]");
z1 = [foo(4)];
z2 = [foo(4)];
List<int> z3 = [foo(4)];
print("baz");
List<dynamic> a0 = baz({3: "hello"});
List<int> a1 = baz({3: "hello"});
List<num> a2 = baz({3: "hello"});
List<num> a3 = baz({3: "hello"} ?? {3: "hello"});
}Currently |
|
And from @lrhn List<T> foo<T>(T x) { print(T); return <T>[x]; }
Object o1 = foo(42);
dynamic o2 = foo(42);
/// And how does FutureOr fit into this?
FutureOr<Object> o3 = foo(42);
FutureOr<dynamic> o3 = foo(42); |
|
Another example is parameters of type dynamic (which can be implicit): T foo<T>(T x) {
print('$T $x');
return x;
}
bar(/*implicit dynamic*/ x, dynamic y, Object z) {}
main() {
bar(foo(1), foo(2), foo(3));
// DDC/DDK print:
// int 1
// int 2
// Object 3
} |
|
We talked about this in person, but I'll capture my thoughts briefly here in case it's helpful. My feeling is that, no, My arguments are:
|
In discussions here: #32291 it came up that strong mode as implemented in the analyzer treats
dynamicin a downwards inference context as an empty context. This is to discuss whether we want this behavior in Dart 2 or not.cc @lrhn @jmesserly @stereotype441 @munificent
The text was updated successfully, but these errors were encountered: