🎉 Successful merge of Copilot and Claude implementations: **Copilot Contributions (Phase 8.3):** - ✅ WASM Box Operations: RefNew/RefGet/RefSet complete implementation - ✅ Memory management: BoxLayout, MemoryManager with standard types - ✅ WASM codegen: Box allocation, field access, type-safe operations - ✅ Runtime support: malloc, heap management, type ID system **Claude Contributions (Benchmark System):** - ✅ Comprehensive benchmark framework (src/benchmarks.rs) - ✅ CLI integration: --benchmark, --iterations, --output options - ✅ 3-backend performance comparison (Interpreter/VM/WASM) - ✅ 280x WASM speedup verification system - ✅ Golden dump testing infrastructure **Unified Features:** - 🔧 execute_wasm_mode: Supports both output file and stdout - 🔧 CLI arguments: All options preserved and functional - 🔧 Error handling: Improved MIR verification messages - 🔧 Build system: All modules properly integrated **Next Steps Ready:** - 📊 MIR diet planning (35→20 instructions) - 🚀 Phase 8.4: AOT WASM native compilation - 🧪 Golden dump automation 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
12 KiB
12 KiB
🏆 Nyash Golden Dump Testing System
ChatGPT5推奨・MIR互換テスト(回帰検出)完全仕様
🎯 目的
「同じ入力→同じ出力」をinterp/vm/wasm/aot間で保証する自動検証システム
MIR仕様の揺れ・バックエンド差異・最適化バグを即座検出し、Portability Contract v0を技術的に保証。
🔧 Golden Dump方式
基本原理
# 1. MIR「黄金標準」生成
nyash --dump-mir program.nyash > program.golden.mir
# 2. 実行時MIR比較(回帰検出)
nyash --dump-mir program.nyash > program.current.mir
diff program.golden.mir program.current.mir
# 3. 全バックエンド出力比較(互換検証)
nyash --target interp program.nyash > interp.out
nyash --target vm program.nyash > vm.out
nyash --target wasm program.nyash > wasm.out
diff interp.out vm.out && diff vm.out wasm.out
階層化検証戦略
| レベル | 検証対象 | 目的 | 頻度 |
|---|---|---|---|
| L1: MIR構造 | AST→MIR変換 | 回帰検出 | 毎commit |
| L2: 実行結果 | stdout/stderr | 互換性 | 毎PR |
| L3: 最適化効果 | 性能・メモリ | 最適化回帰 | 毎週 |
| L4: エラー処理 | 例外・エラー | 堅牢性 | 毎リリース |
🧪 検証テストスイート
1️⃣ MIR Structure Tests (L1)
基本構造検証
// tests/golden_dump/mir_structure_tests.rs
#[test]
fn test_basic_arithmetic_mir_stability() {
let source = r#"
static box Main {
main() {
local a, b, result
a = 42
b = 8
result = a + b
print(result)
return result
}
}
"#;
let golden_mir = load_golden_mir("basic_arithmetic.mir");
let current_mir = compile_to_mir(source);
assert_eq!(golden_mir, current_mir, "MIR回帰検出");
}
#[test]
fn test_box_operations_mir_stability() {
let source = r#"
box DataBox {
init { value }
pack(val) { me.value = val }
}
static box Main {
main() {
local obj = new DataBox(100)
print(obj.value)
}
}
"#;
let golden_mir = load_golden_mir("box_operations.mir");
let current_mir = compile_to_mir(source);
assert_mir_equivalent(golden_mir, current_mir);
}
#[test]
fn test_weak_reference_mir_stability() {
let source = r#"
box Parent { init { child_weak } }
box Child { init { data } }
static box Main {
main() {
local parent = new Parent()
local child = new Child(42)
parent.child_weak = weak(child)
if parent.child_weak.isAlive() {
print(parent.child_weak.get().data)
}
}
}
"#;
verify_mir_golden("weak_reference", source);
}
MIR比較アルゴリズム
// src/testing/mir_comparison.rs
pub fn assert_mir_equivalent(golden: &MirModule, current: &MirModule) {
// 1. 関数数・名前一致
assert_eq!(golden.functions.len(), current.functions.len());
for (name, golden_func) in &golden.functions {
let current_func = current.functions.get(name)
.expect(&format!("関数{}が見つからない", name));
// 2. 基本ブロック構造一致
assert_eq!(golden_func.blocks.len(), current_func.blocks.len());
// 3. 命令列意味的等価性(ValueId正規化)
let golden_normalized = normalize_value_ids(golden_func);
let current_normalized = normalize_value_ids(current_func);
assert_eq!(golden_normalized, current_normalized);
}
}
fn normalize_value_ids(func: &MirFunction) -> MirFunction {
// ValueIdを連番に正規化(%0, %1, %2...)
// 意味的に同じ命令列を確実に比較可能にする
}
2️⃣ Cross-Backend Output Tests (L2)
標準出力一致検証
// tests/golden_dump/output_compatibility_tests.rs
#[test]
fn test_cross_backend_arithmetic_output() {
let program = "arithmetic_test.nyash";
let interp_output = run_backend("interp", program);
let vm_output = run_backend("vm", program);
let wasm_output = run_backend("wasm", program);
assert_eq!(interp_output.stdout, vm_output.stdout);
assert_eq!(vm_output.stdout, wasm_output.stdout);
assert_eq!(interp_output.exit_code, vm_output.exit_code);
assert_eq!(vm_output.exit_code, wasm_output.exit_code);
}
#[test]
fn test_cross_backend_object_lifecycle() {
let program = "object_lifecycle_test.nyash";
let results = run_all_backends(program);
// fini()順序・タイミングが全バックエンドで同一
let finalization_orders: Vec<_> = results.iter()
.map(|r| &r.finalization_order)
.collect();
assert!(finalization_orders.windows(2).all(|w| w[0] == w[1]));
}
#[test]
fn test_cross_backend_weak_reference_behavior() {
let program = "weak_reference_test.nyash";
let results = run_all_backends(program);
// weak参照の生存チェック・null化が同一タイミング
let weak_behaviors: Vec<_> = results.iter()
.map(|r| &r.weak_reference_timeline)
.collect();
assert_all_equivalent(weak_behaviors);
}
エラー処理一致検証
#[test]
fn test_cross_backend_error_handling() {
let error_programs = [
"null_dereference.nyash",
"division_by_zero.nyash",
"weak_reference_after_fini.nyash",
"infinite_recursion.nyash"
];
for program in &error_programs {
let results = run_all_backends(program);
// エラー種別・メッセージが全バックエンドで同一
let error_types: Vec<_> = results.iter()
.map(|r| &r.error_type)
.collect();
assert_all_equivalent(error_types);
}
}
3️⃣ Optimization Effect Tests (L3)
Bus-elision検証
// tests/golden_dump/optimization_tests.rs
#[test]
fn test_bus_elision_output_equivalence() {
let program = "bus_communication_test.nyash";
let elision_on = run_with_flag(program, "--elide-bus");
let elision_off = run_with_flag(program, "--no-elide-bus");
// 出力は同一・性能は差がある
assert_eq!(elision_on.stdout, elision_off.stdout);
assert!(elision_on.execution_time < elision_off.execution_time);
}
#[test]
fn test_pure_function_optimization_equivalence() {
let program = "pure_function_optimization.nyash";
let optimized = run_with_flag(program, "--optimize");
let reference = run_with_flag(program, "--no-optimize");
// 最適化ON/OFFで結果同一
assert_eq!(optimized.output, reference.output);
// PURE関数の呼び出し回数が最適化で削減
assert!(optimized.pure_function_calls <= reference.pure_function_calls);
}
#[test]
fn test_memory_layout_compatibility() {
let program = "memory_intensive_test.nyash";
let results = run_all_backends(program);
// Box構造・フィールドアクセスが全バックエンドで同一結果
let memory_access_patterns: Vec<_> = results.iter()
.map(|r| &r.memory_access_log)
.collect();
assert_memory_semantics_equivalent(memory_access_patterns);
}
性能回帰検証
#[test]
fn test_performance_regression() {
let benchmarks = [
"arithmetic_heavy.nyash",
"object_creation_heavy.nyash",
"weak_reference_heavy.nyash"
];
for benchmark in &benchmarks {
let golden_perf = load_golden_performance(benchmark);
let current_perf = measure_current_performance(benchmark);
// 性能が大幅に劣化していないことを確認
let regression_threshold = 1.2; // 20%まで許容
assert!(current_perf.execution_time <= golden_perf.execution_time * regression_threshold);
assert!(current_perf.memory_usage <= golden_perf.memory_usage * regression_threshold);
}
}
🤖 自動化CI/CD統合
GitHub Actions設定
# .github/workflows/golden_dump_testing.yml
name: Golden Dump Testing
on: [push, pull_request]
jobs:
mir-stability:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v3
- name: Setup Rust
uses: actions-rs/toolchain@v1
with:
toolchain: stable
- name: Run MIR Structure Tests (L1)
run: |
cargo test --test mir_structure_tests
- name: Verify MIR Golden Dumps
run: |
./scripts/verify_mir_golden_dumps.sh
cross-backend-compatibility:
runs-on: ubuntu-latest
needs: mir-stability
steps:
- name: Run Cross-Backend Tests (L2)
run: |
cargo test --test output_compatibility_tests
- name: Verify All Backend Output Equality
run: |
./scripts/verify_backend_compatibility.sh
optimization-regression:
runs-on: ubuntu-latest
needs: cross-backend-compatibility
steps:
- name: Run Optimization Tests (L3)
run: |
cargo test --test optimization_tests
- name: Performance Regression Check
run: |
./scripts/check_performance_regression.sh
自動Golden Dump更新
#!/bin/bash
# scripts/update_golden_dumps.sh
echo "🏆 Golden Dump更新中..."
# 1. 現在のMIRを新しい黄金標準として設定
for test_file in tests/golden_dump/programs/*.nyash; do
program_name=$(basename "$test_file" .nyash)
echo "更新中: $program_name"
# MIR golden dump更新
./target/release/nyash --dump-mir "$test_file" > "tests/golden_dump/mir/${program_name}.golden.mir"
# 出力 golden dump更新
./target/release/nyash --target interp "$test_file" > "tests/golden_dump/output/${program_name}.golden.out"
done
echo "✅ Golden Dump更新完了"
# 2. 更新を確認するためのテスト実行
cargo test --test golden_dump_tests
if [ $? -eq 0 ]; then
echo "🎉 新しいGolden Dumpでテスト成功"
else
echo "❌ 新しいGolden Dumpでテスト失敗"
exit 1
fi
📊 実装優先順位
Phase 8.4(緊急)
- L1実装: MIR構造検証・基本golden dump
- 基本自動化: CI/CDでのMIR回帰検出
- Bus命令テスト: elision ON/OFF検証基盤
Phase 8.5(短期)
- L2実装: 全バックエンド出力一致検証
- エラー処理: 例外・エラーケース検証
- 性能基準: ベンチマーク回帰検出
Phase 9+(中長期)
- L3-L4実装: 最適化・堅牢性検証
- 高度自動化: 自動修復・性能トレンド分析
- 形式検証: 数学的正当性証明
🎯 期待効果
品質保証
- 回帰即座検出: MIR仕様変更のバグを即座発見
- バックエンド信頼性: 全実行環境で同一動作保証
- 最適化安全性: 高速化による動作変更防止
開発効率
- 自動品質確認: 手動テスト不要・CI/CDで自動化
- リファクタリング安全性: 大規模変更の影響範囲特定
- 新機能信頼性: 追加機能が既存動作に影響しない保証
Nyash言語価値
- エンタープライズ品質: 厳密な品質保証プロセス
- 技術的差別化: 「全バックエンド互換保証」の実証
- 拡張性基盤: 新バックエンド追加時の品質維持
📚 関連ドキュメント
- MIRリファレンス: mir-reference.md
- 互換性契約: portability-contract.md
- ベンチマークシステム: ../../../benchmarks/README.md
- CI/CD設定: ../../../.github/workflows/
最終更新: 2025-08-14 - ChatGPT5推奨3点セット完成
Golden Dump Testing = Nyash品質保証の技術的基盤