diagnostic for dynamic?

[ghost]

@davidmorgan commented on Mar 2, 2021, 11:47 AM UTC:

I'm surprised to find that dynamic? is allowed and we even see a few uses in google3 ...

Could we lint that you can always switch it to just dynamic? :)

This issue was moved by pq from dart-lang/linter#2493.

[ghost]

@pq commented on Mar 2, 2021, 3:05 PM UTC:

Or perhaps this should be an analyzer diagnostic? (Would anyone ever want to opt-out?)

/fyi @srawlins @bwilkerson

[ghost]

@srawlins commented on Mar 2, 2021, 3:21 PM UTC:

@leafpetersen @munificent is dynamic? different from dynamic in any way? It'd be awkward to immediately default report against using something specified in the spec...

[ghost]

@srawlins commented on Mar 2, 2021, 3:23 PM UTC:

Currently the analyzer even allows dynamic dispatch on dynamic?:

void f(dynamic? a) {
  print(a + 2);
}

[ghost]

@bwilkerson commented on Mar 2, 2021, 4:40 PM UTC:

There's no difference between dynamic and dynamic?, so there's no value in allowing it in code. I doubt that there would be any objections to making it a hint, especially given that hints can be ignored if necessary.

[ghost]

@leafpetersen commented on Mar 2, 2021, 9:06 PM UTC:

Yeah, dynamic? and void? are completely useless. Never? could be hinted to be equivalent to Null. Null? is also useless. FutureOr<T>? is useless if T is nullable, but you could argue that it's useful documentation.

[ghost]

@srawlins commented on Mar 2, 2021, 9:08 PM UTC:

Thanks for the responses! I agree that it should be Hinted then; dynamic?, void?, etc. would look to me like I'm supposed to read some sort of signal from that ? and I would carefully tiptoe around it for a few minutes before realizing it is meaningless.

[srawlins]

Oof, so many email notifications. Please use "Transfer issue" when transfering an issue from one repo to another, not "Move issue."

[srawlins]

Yeah, dynamic? and void? are completely useless. ... Null? is also useless.

Actually, void? appears to be a parse error.

I'm going to submit a diagnostic for dynamic? and Null?.

Never? could be hinted to be equivalent to Null.

I'm going to leave this alone for now. I don't see people writing Never? hardly ever, and I think the guidance is not as strong.

FutureOr<T>? is useless if T is nullable, but you could argue that it's useful documentation.

If that can be argued, then I think we should let this one be. :)