CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.2.2] - 2026-05-19

Core Architecture Overhaul

Style IR — Language-Neutral Style Facts (NEW)

Introduced a language-independent intermediate representation (src/style_ir/mod.rs) that decouples AST parsing from signal detection:

• StyleIr — Unified struct holding all extracted style facts (panic calls, naming violations, deep nesting, debug calls, excessive params, unsafe blocks, magic numbers, commented-out lines, TODO markers, plus language-specific issues for Go/Python/Java/Ruby/C/TS)
• StyleIrSummary — Stable JSON-serializable summary exposing 25+ metrics including god_function_count, over_engineering_count, code_smell_count, and is_clean_signal_baseline
• StyleIrThresholdSummary — Exposes detection thresholds (excessive_param_threshold: 5, god_function_line_threshold: 50)
• Integrated into all output formats (terminal text, JSON, Markdown, GitHub Actions) for consistent downstream consumption

StyleFinding — Unified Finding Model (NEW)

Standardized observation model (src/finding.rs, 816 lines) replacing ad-hoc issue handling:

• StyleFinding — Structured finding with FindingId, CodeLocation (file/line/column/span/symbol), RuleMeta (name/category/intent), StyleSignal, Severity, Confidence (Low/Medium/High), Evidence (snippet/metric/context), and optional StyleSuggestion
• StyleCategory — 8 categories: Naming, Complexity, Duplication, Comments, DebuggingLeftovers, Structure, Consistency, DependencyStyle
• RuleIntent — 5 intents: Readability, Maintainability, TeamConvention, NoiseReduction, CognitiveLoad
• Bidirectional conversion between StyleFinding ↔ CodeIssue for backward compatibility
• compute_signal_scores_from_findings() — Language-agnostic signal scoring via .signal field (no classify_rule() dependency)
• 40+ unit tests verifying ID determinism, category mapping, confidence scoring, and signal aggregation

SignalDetector — Direct Signal Detection (NEW)

Signal-based analysis layer (src/signals.rs, src/detectors.rs) aggregating 10 behavioral dimensions:

• StyleSignal enum — 10 signals: Duplication, PanicAddiction, NamingChaos, NestedHell, HotfixCulture, OverEngineering, CodeSmells, LegacyCode, TodoMountain, LineCountSmell
• SignalDetector trait — count_violations(), count_violations_with_ir() (avoids redundant parsing), detect_findings(), detect_findings_with_ir()
• 10 detectors: PanicAddictionDetector, NamingChaosDetector, NestedHellDetector, HotfixCultureDetector, OverEngineeringDetector, CodeSmellsDetector, DuplicationDetector, LegacyCodeDetector, TodoMountainDetector, LineCountSmellDetector
• LanguageCapabilityMatrix — Tracks which signals are detectable per language (all 11 grammar languages support all 10 signals)
• aggregate_detector_scores() — Aggregates scores across files with density normalization
• violations_to_score() — log2 density formula capped at 25.0
• classify_rule() — Maps 50+ rule names to their behavioral signal
• Test-file awareness — Per-detector skips_test_files() + config-level [signals] skip-tests = double-layer filtering
• 100+ unit tests across all detectors

LanguageAdapter — Unified Multi-Language API (NEW)

Replaced per-language rule files with a uniform adapter trait (src/language/adapter/, 3000+ lines):

• LanguageAdapter trait — extract_functions(), count_panic_calls(), count_naming_violations(), count_deeply_nested_blocks(), count_debug_calls(), count_excessive_params(), count_unsafe_blocks(), count_magic_numbers(), count_commented_out_code(), count_todo_markers(), count_goroutine_spawns(), count_defer_in_loop(), count_go_convention_violations(), count_python_issues(), count_java_issues(), count_ruby_issues(), count_c_issues(), count_ts_issues(), count_dead_code(), count_duplicate_imports(), has_test_nodes()
• 11 adapters: Rust, Python, JavaScript, TypeScript, Go, Java, Ruby, C, C++, Swift, Zig
• Batch query optimization — query_patterns() + batch_captures() + compute_all() runs all tree-sitter queries in a single cursor traversal per file, reducing AST traversals from 15 to 1
• Language-specific rules: Hungarian notation, abbreviation abuse, naming conventions per language
• Every adapter has 10+ unit tests

StyleProfile — Unified Personality Inference (NEW)

Replaced ad-hoc personality inference with data-driven profiling:

• StyleProfile::from_signal_counts() — Normalizes raw signal counts into 0-25 scores
• StyleProfile::from_signal_scores() — Builds from direct detector scores
• infer_personality_type() — 7 personality archetypes depending on dominant signal combination:
  • The Copy-Paste Artist (high Duplication)
  • The YOLO Engineer (high PanicAddiction)
  • The Trait Wizard (high NestedHell)
  • The Legacy Necromancer (high NamingChaos)
  • The Hotfix Mercenary (high HotfixCulture)
  • The Startup Survivor (med Duplication + med PanicAddiction)
  • The Academic Wizard (high OverEngineering / med NamingChaos + med NestedHell)
  • The Enterprise Bureaucrat (default — all signals low)
• 14 unit tests covering all personality branches

New Modules

Friend Feedback Layer (src/friend/)

"Friend's Perspective" output system that transforms raw metrics into human-readable insights:

• FriendFeedback — Combines FriendMood (Proud/Concerned/Sarcastic/Alarmed/Exhausted), BehaviorPattern (top 3 signals), NextAction (top 3 quick wins)
• FriendMood::from_score() — Maps total score (0-100) to emotional response with emoji
• BehaviorPattern::from_signals() — Ranks signals by score, provides English + Chinese descriptions and actionable suggestions
• NextAction::from_issues() — Sorts issues by severity, shows top 3 with file:line locations
• Bilingual output — print() (English) + print_zh() (Chinese)
• Integrated into terminal output and Markdown report
• 8 unit tests covering mood thresholds, pattern ranking, and construction

Roast Generator (src/treesitter/roast.rs)

• RoastGenerator — Produces sarcastic "friend-like" messages for each StyleCategory
• Per-category roast templates with randomized selection

Reporter Translations (src/reporter/translations.rs)

• Extracted 871 lines of bilingual (en-US / zh-CN) translation tables from display.rs
• Covers personality types, threat levels, diagnostic messages, and code metric labels

Reporter Autopsy (src/reporter/autopsy.rs)

• Standalone autopsy module for per-file deep-dive analysis
• Utilizes StyleProfile for personality inference from signal scores

Helpers & CLI Args (src/helpers.rs, src/args.rs)

• args.rs (304 lines) — Dedicated CLI argument definitions moved from main.rs
• helpers.rs (548 lines) — Shared utility functions: derive_count_score(), derive_radar_score(), derive_danger_zone_score(), quick_scan_score(), overall_score(), parse_date_to_timestamp(), calculate_metrics()

Detectors Registry (src/detectors.rs)

• 10 SignalDetector implementations all operating on StyleIr for language-agnostic detection
• LegacyCodeDetector and TodoMountainDetector explicitly set skips_test_files: false to always scan test code
• LineCountSmellDetector applies per-context thresholds (1000 lines normal, 2000 lines test)

Multi-Language Expansion

Swift & Zig Support (NEW)

• SwiftAdapter — extract_functions, count_panic_calls (force-unwrap !, fatalError, preconditionFailure), count_naming_violations, count_deeply_nested_blocks, count_debug_calls (print, debugPrint, dump)
• ZigAdapter — extract_functions, count_panic_calls (unreachable, @panic, catch unreachable), count_naming_violations, count_deeply_nested_blocks, count_debug_calls (@import("std").debug.print, std.log)
• All detectors extended to support 11 languages (Rust/Python/JS/TS/Go/Java/Ruby/C/C++/Swift/Zig)

Analysis Pipeline

Analyzer (src/analyzer.rs)

• 3-Phase pipeline: Phase 1 = parse & cache ParsedFile, Phase 2 = SignalDetector + StyleIr, Phase 3 = duplication detection
• FullAnalysisResult — Contains issues, file_infos (per-file StyleIr summaries), direct_scores (signal scores), duplication_findings
• StyleIrFileInfo — Per-file metadata with style_ir_summary for JSON output
• is_generated_file() — Comprehensive regex-based filtering (protobuf, node_modules, vendor, .pb., .generated.)
• collect_source_files() — Unified file gathering with path exclusion integration
• Optimization: Shared ParsedFile cache avoids redundant re-parsing across detector phases

Scoring (src/scoring.rs)

• Two-tier scoring: n_score (nuclear severity, log-scaled capped at 40) + d_score (issue density per 1k lines, log-scaled capped at 60) = total 0-100
• calculate_score_with_direct() — Integrates direct signal scores into composite quality score
• CodeQualityScore.signal_scores — New field storing per-signal scores for downstream consumption
• Fixed: Score was 100.0 (Terrible) when no issues → now correctly 0.0 (Excellent)
• Extended test suite for monotonicity, threshold boundaries, zero-division safety

Output & Reporting

Terminal Output

• Integrated FriendFeedback display after score summary
• StyleIrSummary shown in default/brief/summary modes (function count, god functions, code smells, over-engineering)
• Personality type shown in summary
• Bilingual mode: LANG=zh_CN.UTF-8 triggers Chinese roast + translations

JSON Output (--format json)

• New schema with schema_version, issues[], files[] (per-file style_ir_summary), summary.file_count/issue_count/style_ir_summary
• StyleIrSummary includes thresholds for downstream tooling
• VSCode extension updated to parse new JSON schema (normalizeAnalyzeJson())

Markdown Output (--format markdown)

• Complete report with score, personality, signal distribution table, per-file metrics, FriendFeedback section
• Bilingual Markdown header/footer

GitHub Actions (--format github-actions)

• Native GitHub Annotation output for PR comments
• Integrated into CI workflow

Performance

• 18x analysis speedup for large projects — by consolidating multiple tree-sitter traversals into single LanguageAdapter calls
• 10x test execution speedup — via shared_engine (OnceLock) caching tree-sitter grammars across all tests
• Tree-sitter QUERY_CACHE — Thread-local query caching avoids redundant compilation of repeated patterns
• Batch captures — collect_captures_multi() executes merged multi-pattern queries in a single cursor pass
• Parallel analysis — rayon-based parallel file processing with shared engine
• Benchmark suite (benches/performance_tests.rs) with 15 benchmarks across 8 groups:
  • Near-linear scalability: 1 file = 0.81ms, 50 files = 39.79ms (~0.80ms/file constant)
  • Garbage code ~1.6× slower than clean code (stable ratio)
  • Cross-language projects: 4 languages × 4 files = 107ms
  • Full pipeline overhead: analyze_full 61% slower than analyze_file

Accuracy & Quality

• 25 real-world projects tested across 7 languages, ~5.5M lines of code
• Structural rules (deep-nesting, god-function, long-function, code-duplication): 100% TP rate — production-ready for CI gates
• Language-specific rules (must-use, unwrap, abbrev): ~100% TP
• Overall accuracy report published: ACCURACY_REPORT.md
• Per-language accuracy docs: docs/en/ and docs/zh/ directories with full TP/FP sampling (2-3 samples per rule per language)
• Style rules (magic-number, naming, println): ~15-25% TP — identified as advisory-only, needs per-project config tuning
• Context-aware analysis: FileContext system adjusts rule sensitivity for Business/Example/Test/Benchmark/Documentation code
• Magic-number per-language default allowlists (0, 1, 2, -1 excluded globally)

Documentation

• Full bilingual documentation site: docs/en/ and docs/zh/ covering configuration, rules, tools, and accuracy for all 11 languages
• Configuration reference: docs/config-reference.md with example config at docs/examples/basic-config.toml
• Architecture report: OPTIMIZATION_ANALYSIS_UPDATED.md with 5-phase architecture roadmap
• Benchmark report: BENCHMARK_REPORT.md with detailed performance profiles
• Comprehensive audit: todolist.md tracking all findings (H1-H10, M1-M12, L1-L8)
• Code review findings: REVIEW_FINDINGS.md and docs/code_review_2026-05-19.md
• Bootstrap test report: BOOTSTRAP_TEST_REPORT_v0.2.2.md (796 tests, 0 failures)
• README updated with architecture diagram and tool listing

CI/CD & Dev Experience

• GitHub Actions CI/CD: Automated pipeline with -D warnings clippy policy, multi-platform builds (ubuntu/macos/windows), and parallel testing
• GitHub Actions PR Review: Automatic PR comment with roast analysis
• VSCode Extension: Updated to parse new JSON schema + 30-second timeout to prevent hanging
• Makefile: make ci = clippy + test + bench, make doc for docs site, make format
• Pre-push hook: Runs clippy + tests before push via hooks/pre-push
• Deny.toml: License + advisory checks for dependency audit (cargo deny)

Bug Fixes

🔴 Critical

• UTF-8 panic in truncate() — 3 copies in decay/mod.rs, team_roast/mod.rs, commit_roaster/report.rs would panic on multi-byte chars (Chinese, emoji). Extracted shared UTF-8-safe implementation into utils.rs
• Scoring logic bug — Score was 100.0 (Terrible) when no issues → now correctly 0.0 (Excellent)
• VSCode extension hanging — Added 30-second timeout to exec() calls
• LongFunctionRule {} placeholder — Never substituted, always output raw template
• RustMustUseRule duplicate condition — Copy-paste error causing duplicate detection
• RustMustUseRule multi-line signatures — Missed multi-line fn declarations

🔴 High

• Missing Swift/Zig language detection — Language::from_extension() never matched "swift" or "zig", silently classifying as Unknown
• CrossFileAnalyzer error handling — Replaced let _ = with proper error logging
• Example file detection logic — Fixed /messages/ check placement to prevent misclassification
• 5 tree-sitter rules never registered — TerribleNamingRule, SingleLetterTsRule, DeepNestingRule, PrintlnDebuggingRule, MagicNumberRule
• Unwrap-abuse rule never registered for Rust
• 9 language-specific rules never registered: TooManyParamsRule, HungarianNotationTsRule, AbbreviationAbuseTsRule, RustDeriveOrderRule, RustErrorDisplayRule, RustMustUseRule, RustDocExampleRule, CrossFileDuplication, FeatureEnvyRule
• Signal test assertions stale — sigs.len() == 7 should be 10 (missing LegacyCode, TodoMountain, LineCountSmell)
• direct_signals() missing LegacyCode, TodoMountain, LineCountSmell per language

🟡 Medium

• Unclamped score functions — derive_radar_score() and derive_danger_zone_score() did not clamp to [0.0, 100.0]
• Hardcoded Chinese in English mode — get_category_roast() always used Chinese names ("命名规范")
• CLI short flag conflict — team-roast -l used by both --limit and --lang. Changed --lang to -L
• CLI default value ambiguity — llm_timeout changed from u64 to Option<u64>
• 0-param functions counted as 1 — All 11 adapters count_excessive_params() bug
• RustErrorDisplayRule hardcoded line 1 — Now handles generic impls (impl<T> Debug for Foo<T>)
• RustDeriveOrderRule misses multi-line derives — Now handles #[derive(\n Debug, Clone\n)]
• Java has_annotation wrong line — Now checks the line above the method
• TypeScript prefer-interface — Now only flags object types, not unions/primitives
• Python .format() matches strings/comments — Now skips string literals and comments
• Go debug_calls double-counts panic() — Filtered out panic from debug count
• Magic number 1_000_000 parse failure — Handles Rust underscored literals
• is_clean_signal_baseline() — Was missing commented_out_lines and todo_count checks
• Go skip list dead code — Multi-char entries in len==1 filter
• classify_rule() implicit fallback to CodeSmells — magic-number, unnecessary-clone, etc. all fall through
• Silent config parse failure — No user feedback on malformed config

🟢 Low

• Dead code cleanup — Removed unused functions, variables, modules across 15 files
• Module declaration deduplication — main.rs now imports from lib.rs instead of redeclaring modules (faster builds)
• Removed 10 #[allow(dead_code)] annotations — Zero dead code policy enforced
• Deduplicated BLOCK_PARENT_TYPES — Was duplicated across multiple rule files
• Deduplicated direct_signals() — All language arms now share a single array
• Gated slow integration tests behind GCH_INTEGRATION env var
• Go skip list cleanup — Removed dead entries
• Regex recompilation — Switched to LazyLock for per-file regex patterns

[0.2.1]

Added

• Cross-file duplication detection: New ability to detect duplicated code across multiple files using function fingerprinting and similarity matching
• Context-aware analysis: FileContext system that adjusts rule sensitivity based on code type (Business, Example, Test, Benchmark, Documentation)
• GitHub Actions CI/CD: Automated pipeline with 0 warning policy, testing, and multi-platform builds
• Phase 1.3 rule adaptation: Core rules now use check_with_context() method for context-aware analysis:
  • MeaninglessNamingRule: Skips/reduces sensitivity in Test/Example contexts
  • MagicNumberRule: Filters UI-related values in Business context
  • PrintlnDebuggingRule: Completely skips Test/Example files
  • UnwrapAbuseRule: Higher threshold in Test contexts
  • TerribleNamingRule: Skips Example/Demo files
• Shared truncate() utility: UTF-8 safe string truncation in utils.rs, replacing 3 duplicate buggy implementations

Fixed

• 🔴 Critical: UTF-8 panic in truncate() — 3 copies of the same bug in decay/mod.rs, team_roast/mod.rs, commit_roaster/report.rs would panic on multi-byte characters (Chinese, emoji). Extracted shared UTF-8-safe implementation into utils.rs
• 🔴 High: Missing Swift/Zig language detection — Language::from_extension() never matched "swift" or "zig", silently classifying these files as Unknown despite full tree-sitter grammar support
• 🟡 Medium: Unclamped score functions — derive_radar_score() and derive_danger_zone_score() in main.rs did not clamp return values to [0.0, 100.0], unlike derive_count_score()
• 🟡 Medium: Hardcoded Chinese in English mode — reporter/display.rs get_category_roast() always used Chinese category names ("命名规范") regardless of locale setting
• 🟡 Medium: CLI short flag conflict — team-roast subcommand had -l used by both --limit and --lang, causing panic on any invocation. Changed --lang to -L
• 🟢 Low: Dead code cleanup
  • Removed unused all_rules(), single_letter_variable(), deep_nesting() from treesitter/rules/mod.rs
  • Removed unused _max_per_rule variable from reporter/mod.rs
  • Removed unused CommitInfo::hash field from decay/mod.rs
  • Removed dead parse_python() test helper from treesitter/duplication.rs
  • Removed empty src/evolution_chat/ directory
  • Deleted RELEASE_NOTE.md (content merged into CHANGELOG)

Fixed

• 🔴 Critical: Scoring logic bug (scoring.rs:91)

  • Fixed: Score was 100.0 (Terrible) when no issues, should be 0.0 (Excellent)
  • Impact: All clean projects now correctly show score of 0.0

• 🔴 Critical: VSCode extension hanging (extension.ts:462)

  • Added: 30-second timeout to exec() calls to prevent extension freeze when CLI hangs

• 🔴 High: CrossFileAnalyzer error handling (analyzer.rs:110)

  • Changed: Replaced let _ = with proper error logging for failed file processing

• 🔴 High: Example file detection logic (file_context.rs:108-112)

  • Fixed: Moved /messages/ check inside condition block to prevent misclassification

• 🟡 Medium: CLI default value ambiguity (config.rs:137)

  • Changed: llm_timeout parameter from u64 to Option<u64> to distinguish "not set" from "explicitly set to 30"

• 🟡 Medium: Dead code removal

  • Deleted: i18n::get_suggestions() method that always returned empty vec
  • Removed: Unused _rule_count variable in main.rs
  • Cleaned up: Empty inline decoration handlers in VSCode extension
  • Removed: Empty onDidChangeActiveTextEditor event handler

• 🟢 Low: Module declaration deduplication (main.rs, lib.rs)

  • Changed: main.rs now imports from lib.rs instead of redeclaring all modules
  • Impact: Eliminates duplicate compilation, faster build times

Changed

• Removed 10 #[allow(dead_code)] annotations: Project now enforces zero dead code policy
• Improved error messages: Better user feedback for cross-file analysis failures
• Code quality: All public APIs in cross_file module are now properly used or marked internal

[0.1.3] - 2026-05-09

Added

• Educational advice system with bilingual support (English/Chinese)
• Hall of Shame feature for tracking worst offenders
• LLM integration support (Ollama, OpenAI-compatible)
• Markdown output format option
• Verbose mode with detailed diagnostics

Fixed

• Improved println! detection accuracy (reduced false positives)
• Enhanced magic number detection patterns
• Better unwrap abuse warnings

[0.1.2] - 2026-05-08

Added

• Initial VSCode extension support
• Basic rule engine with 25+ rules
• File exclusion patterns
• Multi-language roast messages

[0.1.1] - 2026-05-07

Added

• First working version
• Core analyzer implementation
• Basic scoring system
• CLI interface with clap