dangi_doctor 0.2.4

Flutter app health CLI — crawls every screen, analyses widget trees, measures performance, detects bugs, and generates integration tests.

0.2.4

• Updated README "What's new" section to reflect v0.2.3 changes.

0.2.3

• AI pooling for large screens: when a request exceeds the provider's token limit (413), issues are split into batches of 40 and analysed in parallel passes with a compact system prompt. A final summarisation call combines all batch findings into a unified health report.
• Auto-retry on rate limits (429): parses the "try again in Xs" wait time from Groq / OpenAI error responses and retries automatically — no manual re-runs needed.
• Groq model switched to llama-3.1-8b-instant (30 000 TPM free tier vs 12 000 for 70B), making Groq reliably usable without upgrading.
• Stable tappables polling: crawler now waits for the tappable-element count to stabilise before exploring a screen, catching async-loaded content (API-driven lists, lazy widgets).
• Smart post-tap waiting: replaced fixed 1 200 ms delay with a poll-until-screen-changes strategy — faster on quick navigations, patient on slow ones.

0.2.2

• Expanded Layer 1 AI knowledge with 5 new flutter.dev sources:
  • Layout constraints (flutter.dev/ui/layout/constraints)
  • DevTools performance profiling (flutter.dev/tools/devtools/performance)
  • DevTools memory & leak detection (flutter.dev/tools/devtools/memory)
  • Android deployment guide (flutter.dev/deployment/android)
  • Networking & async cookbook (flutter.dev/cookbook/networking/fetch-data)
• AI diagnosis now free for everyone: added Groq (free tier, no credit card) and Ollama (fully local, no account) to README with setup instructions.
• Knowledge auto-updates every Monday via GitHub Actions.

0.2.1

• Fixed pub.dev score issues:
  • Shortened pubspec.yaml description to fit 60–180 character limit.
  • Updated repository, homepage, and issue_tracker to the correct GitHub URL.
  • Bumped vm_service dependency to ^15.0.0 (latest compatible version).
  • Added example/main.dart (canonical example file pub.dev looks for).
  • Fixed HTML angle brackets in dartdoc (GlobalKey<NavigatorState> → backtick-quoted).
  • Added dartdoc comments to DiscoveredScreen fields, ScreenNavigator fields, and HtmlReportGenerator to push API documentation coverage above 20%.

0.2.0

• Phase 2 navigation completely rewritten — replaced nav-bar heuristic detection with universal tappable exploration. Uses adb shell uiautomator dump to get exact pixel bounds for every clickable element; taps each one and detects new screens by root widget type change or significant source-file delta. Works on any Flutter app with no hardcoded assumptions.
• ListView deduplication: elements with multi-line data labels (e.g. 
 entities) are grouped by screen column; only the first item per group is tapped, preventing wasted taps on long ListView.builder lists.
• Depth-aware back navigation: navDepth tracks forward push steps from home; back-presses are capped at navDepth + 2 so the crawler never overshoots past the target screen.
• Overshoot detection: _returnToScreen stops immediately if it lands on the home screen before reaching the intended target (fixes cascade failures from tab-switch navigation).
• Back-button filtering now catches unlabelled AppBar icons — elements at top-left corner (cx < 200, cy < 380) with no text or auto-generated tap(cx,cy) label are skipped.
• Explored paths persisted to .dangi_doctor/explored_paths.json; subsequent runs offer continue-from-last-run or restart-fresh options.
• Fixed broken example/ file (was importing nonexistent package:cli/cli.dart).
• Fixed all dart analyze warnings; dart format applied throughout.
• lib/cli.dart now exports the real public API (DiscoveredScreen, ScreenNavigator, AppAnalysis, AppAnalyser).

0.1.0

• Initial release.
• Live Flutter app crawler via Dart VM service WebSocket.
• Widget tree analyser: deep nesting, Provider overload, anti-patterns with exact file:line.
• Performance measurement via VM Timeline API (grade A–F per screen).
• Static bug detectors: setState after dispose, stream subscription leaks, build side-effects, late field double-init.
• AI diagnosis (Claude, OpenAI, Gemini, Groq, Ollama) with 3-layer knowledge system.
• Integration test generator: smoke, interaction, and performance test files per screen.
• HTML health report auto-opened in browser after every run.
• Port remapping via /proc/net/tcp — handles Flutter's incorrect observatory port reporting on Android.
• Project fingerprint (Layer 3) generated on first run, saved to .dangi_doctor/project.json.