fix: cap dart_style below 3.1.10 so generated code matches dart format (v2.8.1)

[gmpassos]

Problem

The generated *.reflection.g.dart is formatted by DartFormatter, using the dart_style that pub resolves for this package, while consumers check that same code with the dart format bundled in their Dart SDK.

Those two are not the same version, and dart_style does not guarantee identical output across its own versions.

• Dart SDK 3.12.2 bundles dart_style 3.1.6 (dart format --version → 3.1.6)
• reflection_factory 2.8.0 requires dart_style: ^3.1.9 → pub resolves 3.1.12
• dart_style 3.1.10 changed the splitting of an argument list containing a collection literal

That shape is exactly the generated proxy code:

// dart_style 3.1.12 (what the builder writes) — hugs:
var ret = onCall(this, 'mapKeys', <String, dynamic>{
  'map': map,
}, const __TR<Future<List<String>>>(Future, <__TR>[__TR.tListString]));

// dart_style 3.1.6 (what the consumer's `dart format` wants) — splits:
var ret = onCall(
  this,
  'mapKeys',
  <String, dynamic>{'map': map},
  const __TR<Future<List<String>>>(Future, <__TR>[__TR.tListString]),
);

Why it breaks consumers

It makes two ordinary CI checks mutually exclusive:

Committed content	dart format --set-exit-if-changed	build_verify (build is up to date)
Generator output	❌ fails	✅ passes
dart format output	✅ passes	❌ fails — build_runner rewrites it back

Hit in bones_api, where CI's dart pub upgrade picked up 2.8.0 and the test_vm job started failing on ensure_build_test.

Fix

dart_style: ^3.1.9 → '>=3.1.9 <3.1.10'

This is the narrowest range that still satisfies analyzer: ^13.0.0 — 3.1.9 is the first dart_style that accepts analyzer 13 — so nothing else is rolled back (analyzer stays at 13.3.0).

What this is not

Not a language-version bug. The builder already resolves and passes the correct language version to DartFormatter:

• reflection_factory_builder.dart:301 → codeTable.resolveLanguageVersion(), single _writeGeneratedCode call site, so the ?? DartFormatter.latestLanguageVersion fallback is not reached
• Verified against a real 3.7 package: logs Resolved languageVersion: 3.7.0 for every generated file

And raising the consumer's SDK constraint does not avoid it. Regenerating a real package's *.g.dart with dart_style 3.1.12 at each language version:

Package sdk: lower bound	dart_style	SDK dart format
>=3.7.0	3.1.12	❌ DIVERGE
>=3.10.0	3.1.12	❌ DIVERGE
>=3.11.0	3.1.12	❌ DIVERGE
>=3.12.0	3.1.12	❌ DIVERGE

Tests

New test/reflection_factory_format_test.dart, tagged build:

1. The committed *.g.dart are left untouched by the SDK's dart format.
2. DartFormatter agrees with the SDK's dart format on an argument list containing a collection literal — the generated-proxy shape — formatted at this package's own language version.

Test 2 is the one that actually catches this. Test 1 alone is not sufficient: this repository's own generated code happens not to contain a diverging shape, so it passes even on dart_style 3.1.12 while a downstream package with a proxy breaks. Verified by running both against 3.1.12 — only test 2 fails, with:

`DartFormatter` (`dart_style` 3.1.12) and the `dart format` of the Dart SDK (3.1.6)
format the same code differently, at language version 3.10.0.

Full suite: 272 tests pass; dart format and dart analyze --fatal-infos --fatal-warnings clean.

Follow-up

Lift the upper bound once the Dart SDK bundles dart_style >= 3.1.10 — at that point test 2 keeps the two aligned automatically.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.51%. Comparing base (53a8c3a) to head (a24e40b).

Additional details and impacted files

@@           Coverage Diff           @@
##           master      #51   +/-   ##
=======================================
  Coverage   89.51%   89.51%           
=======================================
  Files          14       14           
  Lines        6398     6398           
=======================================
  Hits         5727     5727           
  Misses        671      671           

Flag	Coverage Δ
unittests	89.51% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.