[cfe] Instantiated function tearoff does have Object methods

[eernstg]

[Edit] We do have a rule here which says that a term like a<b, c>.toString() must be parsed as a constructor invocation when a<b, c> is a type literal, not as an instance method invocation on that type literal. But the specification does not explicitly address the case where a<b, c> is a generic function instantiation.

Old text (striking out the part which is currently not resolved, cf. dart-lang/language#1802):

Consider the test co19/LanguageFeatures/Constructor-tear-offs/ambiguities_A08_t02.dart.

In this test there is an expression a<b, c>.toString() where a is a generic function and b and c are types. The parser should parse a<b, c> as <primary><selector> where <selector> further derives <typeArguments>, because the next token is '.' (which is one of those tokens that force the preceding construct to be considered as a <typeArguments>). I believe this part works as it should.

However, that expression should then denote a toString() invocation on a function object that is obtained by explicit generic function instantiation, and there should not be any errors. The CFE reports a compile-time error Method not found.

[eernstg]

The same situation seems to exist in co19/LanguageFeatures/Constructor-tear-offs/ambiguities_A08_t03.dart.

[eernstg]

Same situation in co19/LanguageFeatures/Constructor-tear-offs/ambiguities_A13_t03.dart.

[eernstg]

The core question here is currently being resolved at dart-lang/language#1802. Re-opening and marking as blocked.

[stereotype441]

Was just discussing this bug with @srawlins and realized that if we want a<b, c>.toString() to be allowed (where a refers to a generic function/method), then a<b, c>.call(...) should probably be allowed too.

(@eernstg I think you've been working on test cases for this sort of thing? You might want to add some test cases for .call if you haven't already).

[eernstg]

Thanks! a<b, c>.call(...) should surely be allowed whenever a has a function type or a.call is a constructor name, I'll include that.

[lrhn]

(The analyzer might want to hint that a<b, c>(...) is a shorter version of the same thing, though).

[stereotype441]

(The analyzer might want to hint that a<b, c>(...) is a shorter version of the same thing, though).

Good point. @pq or @bwilkerson, do you know whether this would be better as a lint or a hint? And would you mind taking ownership of this feature request?

[eernstg]

dart-lang/language#1802 has been closed: f<int>.toString() is a non-error term that invokes toString() on a function object, and C<int>.toString() is an error unless C.toString is the name of a constructor.