Convert nullable state fields to late or late final, if the fields are initialized in initState.

[polina-c]

For example, for the code

class MyClassState extends State<MyClass> {
  MyClassState();

  String? _name;

  @override
  void initState() {
    super.initState();

    _name = detectName(); 
  }
...

The lint would suggest:

class MyClassState extends State<MyClass> {
  MyClassState();

  late final String _name;
  ...

(we should not do it for fields that are assigned to null in dispose)

[jacob314]

This lint could also have some false positives. Creating lints that don't have some false positives is very trick. I would make the field late as you could consider a separate lint about generally promoting fields to final.
To minimize false positives, the fields would need to be private or the class private (generally the case in State objects) and you would need zero calls to set the field to a possibly nullable value.
Here is an example without a dispose method where the field should not be made late as the token is set to null some of the time in the click handler.

class MyClassState extends State<MyClass> {
  MyClassState();

  Token? _token;

  @override
  void initState() {
    super.initState();

    _token = Token(); 
  }
  
  Token? maybeCreateToken() {...}

  void onClick() {
    _token = MaybeCreateToken();
  }

[pq]

/fyi @goderbauer

[goderbauer]

What is the motivation for the lint? Right now I don't see the immediate value in it and like Jacob I am a little concerned that there is a lot of potential for false positives.

As a side note, for the sample code mentioned in the issue, I would really refactored that particular sample to be even more concise:

class MyClassState extends State<MyClass> {
  late final String _name = detectName();

  // no initState override necessary.
}

But again, I am not convinced that that's a general pattern that needs encoding in a lint.

[jacob314]

I wish we had some way of quickly searching a corpus of Dart code to estimate about how many violations a lint like this would trigger so we could then spot check to see how many of those cases are important.
I'd summarize the motivation as "teach users to use late instead of creating nullable fields". Potentially we could audit nullable fields created by all users and see what % seem like they should be late to understand if lints would have impact here or not.