`abstract_class_member` triggering inside `abstract class`

[FMorschel]

Repro:

abstract class C {
  abstract C operator +(Cother);
//^ abstract_class_member
}

Message:

Members of classes can't be declared to be 'abstract'.
Try removing the 'abstract' keyword. You can add the 'abstract' keyword before the class declaration.

I'm guessing that operators can't be declared abstract, but this message makes no sense.

CC @bwilkerson

[FMorschel]

The same happens with:

abstract class C {
  abstract int get foo;
  abstract set foo(int value);
}

[bwilkerson]

The message means exactly what it says: the modifier abstract can't be used on any members of any class-like declaration.

A member is abstract if it appears in a class-like declaration that is marked as abstract and doesn't have an implementation. (Not having an implementation is an error if the class isn't abstract.) Removing the keyword is the only thing you need to do.

Happy to hear if you have suggestions for how better to word the message.

[FMorschel]

In this case, the class is already abstract. This is what makes the wording confusing.

Maybe something along the lines of:

Members of abstract classes can't be declared to be 'abstract'.
Remove the 'abstract' keyword. You can add an empty body `;` to mean that this member is abstract.

Note that the try removing on the correction is a bit weird since there is no way we can ever make it valid code with that present.

[bwilkerson]

I like the reworded message, but I'd use "Try removing the 'abstract' keyword." for the correction. We have a convention to start every correction message with "Try" because of the tone it sets, and the members already have a semicolon, so there's nothing to add.

I believe that this diagnostic is reported by the parser.