Added #

• WindPerfResolver abstract class (lib/fluttersdk_wind_diagnostics_contracts.dart): a second, separate contract alongside WindDebugResolver, with a single method Map<String, Object?> stats(). Where WindDebugResolver.resolve(Element) resolves per-Element widget state, WindPerfResolver.stats() resolves process-wide performance counters that have no single Element to attach to. The returned map's key set is pinned as the cross-repo contract read by fluttersdk_dusk's performance snapshot, all int: cacheHits, cacheMisses, cacheBypasses, cacheSize, wDivBuilds, wTextBuilds. Additive-growth rule applies: new keys are a minor bump, renaming or removing a key is a major bump.

• WindDebugRegistry perf slot: a second, distinct static slot mirroring the existing debug-resolver triad. WindDebugRegistry.currentPerf (getter, null when unregistered), WindDebugRegistry.registerPerf(resolver) (canonical install path, idempotent, most-recent-call wins). resetForTesting() now clears both slots. Registering a perf resolver never touches the debug slot and vice versa.

• WindDebugRegistry.registerPerfForTesting: the sibling of registerForTesting, and it exists for the same reason. registerPerf is documented as wind's canonical install path, so a consumer's test installing a fake through it reads as production wiring at the call site.

• 6 contract tests (test/fluttersdk_wind_diagnostics_contracts_test.dart): null-when-unregistered, register stores + lookup via stats(), resetForTesting clears the perf slot, registering a perf resolver leaves the debug slot untouched, registerPerf idempotency (most-recent-call wins), and registerPerfForTesting touching only the perf slot.

Changed #

• The coverage gate is no longer balanced on a single line. WindDebugRegistry._() is a private constructor whose only job is to stop this static-only registry being instantiated, so no test can reach it, and it was one of only five executable lines in the package: coverage sat at exactly 80.00% (LH=4, LF=5) against an 80% floor, and one new uncovered line anywhere would have failed CI. It now carries // coverage:ignore-line with the reason above it, which takes the count to 100.00% (LH=4, LF=4). This adds no test; it stops counting a line that cannot be tested. Verified that flutter test --coverage honours the pragma in this toolchain rather than assuming it. (lib/fluttersdk_wind_diagnostics_contracts.dart)

Fixed #

• analysis_options.yaml now carries the analyzer.exclude block that the Flutter tool's migrator writes, so the migrator is a no-op and a CI checkout stays clean. The migrator runs on every flutter pub get, which is the first step of both ci.yml and publish.yml, and flutter pub publish --dry-run later in the same job then reported 1 checked-in file is modified in git and exited 65. Nothing had caught it because the last CI run here predates the migrator (2026-05-21); the next run on any branch would have failed, and the publish workflow shares the step, so it would have blocked a release too. Verified by reproducing the exit 65 and then confirming a second pub get leaves the tree clean with the explanatory comment intact. The same fix landed in fluttersdk_wind as #178.

Initial stable release. Neutral abstract contracts that let debug-tooling packages read Wind UI widget state at runtime without a compile-time dependency on fluttersdk_wind.

Added #

• WindDebugResolver abstract class (lib/fluttersdk_wind_diagnostics_contracts.dart): single-method interface Map<String, Object?> resolve(Element element). Implementations live in fluttersdk_wind (production) and test fakes (debug-only). The returned map's key set is documented as the v1 frozen contract:

  • className: String (required when the widget is a W-prefixed widget with a non-empty className).
  • breakpoint: String (e.g. 'sm', 'md', 'lg').
  • brightness: String ('light' or 'dark').
  • platform: String (e.g. 'web', 'ios', 'android', 'macos').
  • states: List<String> (e.g. ['hover', 'focus']).
  • bgColor: String hex (e.g. '#3B82F6'); present only when resolved.
  • textColor: String hex; present only when resolved. Implementations MUST return const {} for elements that are not Wind widgets, guaranteeing a graceful no-op for consumers walking arbitrary widget trees.

• WindDebugRegistry static registry (lib/fluttersdk_wind_diagnostics_contracts.dart): process-global single-slot registration for the active WindDebugResolver. Wind installs its concrete resolver at app boot (gated by kDebugMode) via Wind.installDebugResolver(); debug-tooling consumers look up the current resolver via WindDebugRegistry.current. The registry exposes:

  • WindDebugRegistry.current getter (returns null when no resolver is registered).
  • WindDebugRegistry.register(resolver) (canonical install path; idempotent, most-recent-call wins).
  • WindDebugRegistry.resetForTesting() and WindDebugRegistry.registerForTesting(resolver) (both @visibleForTesting; test-only seams).

• 6 contract tests (test/fluttersdk_wind_diagnostics_contracts_test.dart): cover null-when-unregistered, register stores + lookup, idempotent register (last-wins), resetForTesting clears, registerForTesting overrides, and the contract assertion that resolver implementations return const {} for non-Wind elements.

• GitHub Actions workflows (.github/workflows/ci.yml, .github/workflows/publish.yml): CI runs flutter pub get + dart format --set-exit-if-changed + dart analyze + flutter test --coverage with an 80% line-coverage floor + flutter pub publish --dry-run. Publish runs on a SemVer tag push, validates, performs the OIDC publish to pub.dev, then creates a GitHub Release whose body is extracted from this CHANGELOG.

Why this package #

Wind UI exposes 6 widget-state fields (className, breakpoint, brightness, platform, states, bgColor, textColor) that debug-tooling packages such as fluttersdk_dusk need to embed in their snapshot output. Shipping that handoff through fluttersdk_wind's own surface would force every debug-tooling package to compile-time depend on Wind (heavy, transitively pulls Flutter's rendering surface), and would force Wind to compile-time depend on every debug-tooling package it wants to feed.

This package breaks the loop. Both sides depend on the abstract contract here; neither side imports the other. The pattern mirrors Flutter's *_platform_interface convention (4.97M downloads on plugin_platform_interface as the canonical precedent).