dart_fmt breaks code with `// ignore: ...` comments at end of line

[zoechi]

I have generated code like

import 'package:fhir_client_api/fhir_client_api.dart' show
  FhirDate, // ignore: unused_shown_name
  FhirXhtml, // ignore: unused_shown_name
  PropertyMetadata, ElementDefinitionBindingMeta, ReferenceMeta, // ignore: unused_shown_name
  ClassMetadata

after dartfmt I get something like

import 'package:fhir_client_api/fhir_client_api.dart' show
  FhirDate, // ignore: unused_shown_name
  FhirXhtml, // ignore: unused_shown_name
  PropertyMetadata, 
  ElementDefinitionBindingMeta, ReferenceMeta, // ignore: unused_shown_name
  ClassMetadata

This shows a warning that PropertyMetadata is shown but not used (according to the linter rule unused_shown_name)

I think lines with // ignore: ... at the end must not be wrapped

[munificent]

I think lines with // ignore: ... at the end must not be wrapped

I don't think we can feasibly do that in the formatter. The decision to split or not may affect many other lines, and the rules encoding those decisions would become magnificently complex if they had to support not splitting on some random subset of lines.

For example, a named argument list either splits before each argument, or before none of them:

// allowed:
someMethod(one: 1, two: 2, three: 3);
someMethod(
    one: 1,
    two: 2,
    three: 3);

// not allowed:
someMethod(
    one: 1, two: 2,
    three: 3);

But if we couldn't split before ignore lines, it would have to handle that, which in turn affects things like the indentation of subexpressions, etc.

A simpler solution is for the user to just copy the // ignore: and add it to the new lines if dartfmt splits something up. In your example, those shows should be each on one line anyway. That's how the formatter formats them.

[zoechi]

It doesn't bother me too much.
Thanks for your explanation.