lint request: ambiguous imports which include an element from the Dart SDK

[srawlins]

Analyzer has a diagnostic for when two imports provide an element of the same name:

// a.dart ////////////////////
class File {}

// b.dart ////////////////////
class File {}

// c.dart ////////////////////
import 'a.dart';
import 'b.dart';

main() {
  File(); // error: ambiguous_import
}

But this error does not appear when a dart: element is at play.

// a.dart ////////////////////
class File {}

// c.dart ////////////////////
import 'a.dart';
import 'dart:io';

main() {
  File(); // MISSING ERROR
}

Flutter is hitting this in a few places, like https://github.com/flutter/flutter/blob/master/packages/flutter_goldens/test/flutter_goldens_test.dart with "package:file/file.dart" and "dart:io". It does not cause runtime problems, but I think is still confusing.

[srawlins]

Maybe I'm missing something about the rules of ambiguous imports, because CFE matches analyzer's behavior. The first example results in this error:

c.dart:7:3: Error: 'File' is imported from both 'a.dart' and 'b.dart'.
  File();
  ^^^^

The second example results in no error.

[srawlins]

@scheglov points out the exception is specified in the spec:

"a declaration from a non-system library shadows declarations from system libraries".

I imagine authors generally still want a check here though, so converting this to a lint request.

[bwilkerson]

The exception was added so that the core libraries could be extended without breaking previously existing code.

Given that, why do you think users would want this lint? What broken code will this help to catch?

[srawlins]

I think most users are unaware of this singular exception to library element name collisions. There are libraries in flutter's codebase which import both dart:io and package:file/file.dart, and reference File. I believe most developers wouldn't know which File is being used in such a library.

If I had files like that in my code, I would want a lint to let me know, so that I could hide names that are being shadowed.

[mateusfccp]

I agree with this. It's unexpected for me that it's not an error, but after the explanation of why this is this way, I can understand why.

Even so, having an opt-in would be interesting, as, most of the time, shadowing a core identifier is done without intention, and may be overlooked.