Optimize SwiftLint: switch to only_rules for 16x speedup

[kj6dev]

Summary

SwiftLint is running 88 "invisible" default-ON rules not explicitly mentioned in our config. Switching from disabled_rules to only_rules yields a 16x performance improvement (1.27s → 0.08s).

Benchmarks

Config	Time	Rules
Current (disabled_rules)	1.27s	107 rules
only_rules (no SourceKit)	0.08s	20 rules
swiftlintcustom-smart	0.07s	14 rules

Root Cause

SwiftLint loads ALL ~200 rules, then applies disabled_rules filter. With only_rules, it only loads what you specify.

Additionally, many enabled rules use SourceKit (compiler queries), which are slow:

• file_length
• statement_position
• trailing_whitespace
• vertical_whitespace
• unused_import (analyzer)

Invisible Default Rules (88 total)

These rules are running but not mentioned in our config:

Formatting/Style (SwiftFormat handles these)

• closing_brace, colon, comma, leading_whitespace
• operator_whitespace, return_arrow_whitespace
• statement_position, trailing_newline, trailing_semicolon
• trailing_whitespace, vertical_whitespace, vertical_parameter_alignment

Redundancy Detection (useful, keep)

• redundant_discardable_let, redundant_objc_attribute
• redundant_optional_initialization, redundant_set_access_control
• redundant_string_enum_value, redundant_void_return, redundant_sendable

Legacy Code Detection (useful, keep)

• legacy_cggeometry_functions, legacy_constant, legacy_constructor
• legacy_hashing, legacy_random, legacy_nsgeometry_functions

Safety/Bugs (useful, keep)

• force_cast, force_try, duplicate_enum_cases, duplicate_imports
• unused_closure_parameter, unused_optional_binding, unused_enumerated
• duplicate_conditions, duplicated_key_in_dictionary_literal

Low Value / Noise (consider dropping)

• block_based_kvo, class_delegate_protocol, xctfail_message
• valid_ibinspectable, private_unit_test, nsobject_prefer_isequal
• inclusive_language, orphaned_doc_comment

Metrics (configured but using SourceKit)

• cyclomatic_complexity, file_length, function_parameter_count
• large_tuple, nesting, type_body_length, type_name

Recommendations

1. Create only_rules config

Replace disabled_rules approach with explicit allowlist:

only_rules:
  # Safety
  - force_cast
  - force_try
  - duplicate_imports
  - duplicate_enum_cases
  - unused_closure_parameter
  
  # Redundancy
  - redundant_discardable_let
  - redundant_optional_initialization
  - redundant_void_return
  
  # Legacy
  - legacy_constructor
  - legacy_random
  
  # Metrics (non-SourceKit)
  - cyclomatic_complexity
  - nesting
  
  # Opt-ins we already use
  - empty_count
  - empty_string
  - first_where
  - last_where
  - toggle_bool
  - implicit_return

2. Remove rules SwiftFormat handles

These are redundant with SwiftFormat:

• All spacing rules (colon, comma, operator_whitespace, etc.)
• trailing_whitespace, trailing_newline
• vertical_whitespace

3. Consider two configs

Configs/
├── shared-swiftlint.yml         # Full config for CI
└── shared-swiftlint-fast.yml    # Fast config for local dev

Add --fast flag to swiftlint-smart for quick iterations.

4. Move SourceKit rules to CI-only

These are valuable but slow:

• unused_import (analyzer rule)
• file_length

Run them in CI, not on every save.

Impact

• 16x faster local linting
• Explicit control over what runs
• No duplicate checks with SwiftFormat
• Same coverage for meaningful rules

Related

• Custom rules already achieve 0.07s by being focused and avoiding SourceKit
• SwiftFormat handles most style/formatting rules better