Flutter Dart Code Review

Flutter/Dart Code Review Best Practices

Comprehensive, library-agnostic checklist for reviewing Flutter/Dart applications. These principles apply regardless of which state management solution, routing library, or DI framework is used.

1. General Project Health

• Project follows consistent folder structure (feature-first or layer-first)
• Proper separation of concerns: UI, business logic, data layers
• No business logic in widgets; widgets are purely presentational
• pubspec.yaml is clean — no unused dependencies, versions pinned appropriately
• analysis_options.yaml includes a strict lint set with strict analyzer settings enabled
• No print() statements in production code — use dart:developer log() or a logging package
• Generated files (.g.dart, .freezed.dart, .gr.dart) are up-to-date or in .gitignore
• Platform-specific code isolated behind abstractions

2. Dart Language Pitfalls

• Implicit dynamic: Missing type annotations leading to dynamic — enable strict-casts, strict-inference, strict-raw-types
• Null safety misuse: Excessive ! (bang operator) instead of proper null checks or Dart 3 pattern matching (if (value case var v?))
• Type promotion failures: Using this.field where local variable promotion would work
• Catching too broadly: catch (e) without on clause; always specify exception types
• Catching Error: Error subtypes indicate bugs and should not be caught
• Unused async: Functions marked async that never await — unnecessary overhead
• late overuse: late used where nullable or constructor initialization would be safer; defers errors to runtime
• String concatenation in loops: Use StringBuffer instead of + for iterative string building
• Mutable state in const contexts: Fields in const constructor classes should not be mutable
• Ignoring Future return values: Use await or explicitly call unawaited() to signal intent
• var where final works: Prefer final for locals and const for compile-time constants
• Relative imports: Use package: imports for consistency
• Mutable collections exposed: Public APIs should return unmodifiable views, not raw List/Map
• Missing Dart 3 pattern matching: Prefer switch expressions and if-case over verbose is checks and manual casting
• Throwaway classes for multiple returns: Use Dart 3 records (String, int) instead of single-use DTOs
• print() in production code: Use dart:developer log() or the project's logging package; print() has no log levels and cannot be filtered

3. Widget Best Practices

Widget decomposition:

• No single widget with a build() method exceeding ~80-100 lines
• Widgets split by encapsulation AND by how they change (rebuild boundaries)
• Private _build*() helper methods that return widgets are extracted to separate widget classes (enables element reuse, const propagation, and framework optimizations)
• Stateless widgets preferred over Stateful where no mutable local state is needed
• Extracted widgets are in separate files when reusable

Const usage:

• const constructors used wherever possible — prevents unnecessary rebuilds
• const literals for collections that don't change (const [], const {})
• Constructor is declared const when all fields are final

Key usage:

• ValueKey used in lists/grids to preserve state across reorders
• GlobalKey used sparingly — only when accessing state across the tree is truly needed
• UniqueKey avoided in build() — it forces rebuild every frame
• ObjectKey used when identity is based on a data object rather than a single value

Theming & design system:

• Colors come from Theme.of(context).colorScheme — no hardcoded Colors.red or hex values
• Text styles come from Theme.of(context).textTheme — no inline TextStyle with raw font sizes
• Dark mode compatibility verified — no assumptions about light background
• Spacing and sizing use consistent design tokens or constants, not magic numbers

Build method complexity:

• No network calls, file I/O, or heavy computation in build()
• No Future.then() or async work in build()
• No subscription creation (.listen()) in build()
• setState() localized to smallest possible subtree

4. State Management (Library-Agnostic)

These principles apply to all Flutter state management solutions (BLoC, Riverpod, Provider, GetX, MobX, Signals, ValueNotifier, etc.).

Architecture:

• Business logic lives outside the widget layer — in a state management component (BLoC, Notifier, Controller, Store, ViewModel, etc.)
• State managers receive dependencies via injection, not by constructing them internally
• A service or repository layer abstracts data sources — widgets and state managers should not call APIs or databases directly
• State managers have a single responsibility — no "god" managers handling unrelated concerns
• Cross-component dependencies follow the solution's conventions:
  • In Riverpod: providers depending on providers via ref.watch is expected — flag only circular or overly tangled chains
  • In BLoC: blocs should not directly depend on other blocs — prefer shared repositories or presentation-layer coordination
  • In other solutions: follow the documented conventions for inter-component communication

Immutability & value equality (for immutable-state solutions: BLoC, Riverpod, Redux):

• State objects are immutable — new instances created via copyWith() or constructors, never mutated in-place
• State classes implement == and hashCode properly (all fields included in comparison)
• Mechanism is consistent across the project — manual override, Equatable, freezed, Dart records, or other
• Collections inside state objects are not exposed as raw mutable List/Map

Reactivity discipline (for reactive-mutation solutions: MobX, GetX, Signals):

• State is only mutated through the solution's reactive API (@action in MobX, .value on signals, .obs in GetX) — direct field mutation bypasses change tracking
• Derived values use the solution's computed mechanism rather than being stored redundantly
• Reactions and disposers are properly cleaned up (ReactionDisposer in MobX, effect cleanup in Signals)

State shape design:

• Mutually exclusive states use sealed types, union variants, or the solution's built-in async state type (e.g. Riverpod's AsyncValue) — not boolean flags (isLoading, isError, hasData)
• Every async operation models loading, success, and error as distinct states
• All state variants are handled exhaustively in UI — no silently ignored cases
• Error states carry error information for display; loading states don't carry stale data
• Nullable data is not used as a loading indicator — states are explicit

// BAD — boolean flag soup allows impossible states
class UserState {
  bool isLoading = false;
  bool hasError = false; // isLoading && hasError is representable!
  User? user;
}

// GOOD (immutable approach) — sealed types make impossible states unrepresentable
sealed class UserState {}
class UserInitial extends UserState {}
class UserLoading extends UserState {}
class UserLoaded extends UserState {
  final User user;
  const UserLoaded(this.user);
}
class UserError extends UserState {
  final String message;
  const UserError(this.message);
}

// GOOD (reactive approach) — observable enum + data, mutations via reactivity API
// enum UserStatus { initial, loading, loaded, error }
// Use your solution's observable/signal to wrap status and data separately

Rebuild optimization:

• State consumer widgets (Builder, Consumer, Observer, Obx, Watch, etc.) scoped as narrow as possible
• Selectors used to rebuild only when specific fields change — not on every state emission
• const widgets used to stop rebuild propagation through the tree
• Computed/derived state is calculated reactively, not stored redundantly

Subscriptions & disposal:

• All manual subscriptions (.listen()) are cancelled in dispose() / close()
• Stream controllers are closed when no longer needed
• Timers are cancelled in disposal lifecycle
• Framework-managed lifecycle is preferred over manual subscription (declarative builders over .listen())
• mounted check before setState in async callbacks
• BuildContext not used after await without checking context.mounted (Flutter 3.7+) — stale context causes crashes
• No navigation, dialogs, or scaffold messages after async gaps without verifying the widget is still mounted
• BuildContext never stored in singletons, state managers, or static fields

Local vs global state:

• Ephemeral UI state (checkbox, slider, animation) uses local state (setState, ValueNotifier)
• Shared state is lifted only as high as needed — not over-globalized
• Feature-scoped state is properly disposed when the feature is no longer active

5. Performance

Unnecessary rebuilds:

• setState() not called at root widget level — localize state changes
• const widgets used to stop rebuild propagation
• RepaintBoundary used around complex subtrees that repaint independently
• AnimatedBuilder child parameter used for subtrees independent of animation

Expensive operations in build():

• No sorting, filtering, or mapping large collections in build() — compute in state management layer
• No regex compilation in build()
• MediaQuery.of(context) usage is specific (e.g., MediaQuery.sizeOf(context))

Image optimization:

• Network images use caching (any caching solution appropriate for the project)
• Appropriate image resolution for target device (no loading 4K images for thumbnails)
• Image.asset with cacheWidth/cacheHeight to decode at display size
• Placeholder and error widgets provided for network images

Lazy loading:

• ListView.builder / GridView.builder used instead of ListView(children: [...]) for large or dynamic lists (concrete constructors are fine for small, static lists)
• Pagination implemented for large data sets
• Deferred loading (deferred as) used for heavy libraries in web builds

Other:

• Opacity widget avoided in animations — use AnimatedOpacity or FadeTransition
• Clipping avoided in animations — pre-clip images
• operator == not overridden on widgets — use const constructors instead
• Intrinsic dimension widgets (IntrinsicHeight, IntrinsicWidth) used sparingly (extra layout pass)

6. Testing

Test types and expectations:

• Unit tests: Cover all business logic (state managers, repositories, utility functions)
• Widget tests: Cover individual widget behavior, interactions, and visual output
• Integration tests: Cover critical user flows end-to-end
• Golden tests: Pixel-perfect comparisons for design-critical UI components

Coverage targets:

• Aim for 80%+ line coverage on business logic
• All state transitions have corresponding tests (loading → success, loading → error, retry, etc.)
• Edge cases tested: empty states, error states, loading states, boundary values

Test isolation:

• External dependencies (API clients, databases, services) are mocked or faked
• Each test file tests exactly one class/unit
• Tests verify behavior, not implementation details
• Stubs define only the behavior needed for each test (minimal stubbing)
• No shared mutable state between test cases

Widget test quality:

• pumpWidget and pump used correctly for async operations
• find.byType, find.text, find.byKey used appropriately
• No flaky tests depending on timing — use pumpAndSettle or explicit pump(Duration)
• Tests run in CI and failures block merges

7. Accessibility

Semantic widgets:

• Semantics widget used to provide screen reader labels where automatic labels are insufficient
• ExcludeSemantics used for purely decorative elements
• MergeSemantics used to combine related widgets into a single accessible element
• Images have semanticLabel property set

Screen reader support:

• All interactive elements are focusable and have meaningful descriptions
• Focus order is logical (follows visual reading order)

Visual accessibility:

• Contrast ratio >= 4.5:1 for text against background
• Tappable targets are at least 48x48 pixels
• Color is not the sole indicator of state (use icons/text alongside)
• Text scales with system font size settings