Files
hakorune/tools/smokes
nyash-codex 376857a81f fix(perf): stabilize MIR emit for ny-llvmc/EXE benchmarks
Problem:
- Stage-B JSON extraction used fragile `awk '/^{/,/^}$/'`
- stdout noise caused empty JSON and bench failures
- arraymap/matmul/maplin --exe mode failed with "failed to emit MIR JSON"

Solution:
- Python3-based robust JSON extraction
  - Search for "kind":"Program" marker
  - Balance braces with quote/escape awareness
  - Resilient to stdout noise
- FORCE jsonfrag mode priority (HAKO_MIR_BUILDER_LOOP_FORCE_JSONFRAG=1)
  - Bypasses Stage-B entirely when set
  - Generates minimal while-form MIR with PHI nodes
- Multi-level fallback strategy
  - L1: Stage-B + selfhost/provider builder
  - L2: --emit-mir-json CLI direct path
  - L3: Minimal jsonfrag MIR generation
- cd $ROOT for Stage-B (fixes using resolution context)

Results:
-  arraymap --exe: ratio=200.00% (was failing)
-  matmul --exe: ratio=200.00% (was failing)
-  maplin --exe: ratio=100.00% (was failing)
-  Existing canaries: aot_prep_e2e_normalize_canary_vm.sh PASS
-  New canary: emit_mir_canary.sh PASS

Known Issues (workarounds applied):
- Stage-B compiler broken (using resolution: StringHelpers.skip_ws/2)
- --emit-mir-json CLI broken (undefined variable: local)
- Current jsonfrag mode bypasses both issues

Documentation:
- benchmarks/README.md: Added MIR emit stabilization notes
- ENV_VARS.md: Already documents HAKO_SELFHOST_BUILDER_FIRST, etc.

Next: Fix Stage-B using resolution to re-enable full optimization path

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
2025-11-13 17:23:48 +09:00
..

Hakorune Smoke Tests v2 — Guide (Nyash-compatible)

Overview

  • Entry: tools/smokes/v2/run.sh — unified runner for quick/integration/full.
  • Profiles:
    • quick — fast developer checks.
    • integration — VM↔LLVM parity, basic stability.
    • full — comprehensive matrix.

🎯 Two Baselines (Runbook)

これから開発の基準となる2つのベースライン

📦 VM ラインRust VM - 既定)

用途: 開発・デバッグ・検証用(高速・型安全)

# ビルド
cargo build --release

# 一括スモークテスト
tools/smokes/v2/run.sh --profile quick

# 個別スモークテスト
tools/smokes/v2/run.sh --profile quick --filter "<glob>"
# 例: --filter "core/json_query_min_vm.sh"

# 単発実行(参考)
$NYASH_BIN --backend vm apps/APP/main.hako

llvmlite ラインLLVMハーネス

用途: 本番・最適化・配布用(実証済み安定性)

前提: Python3 + llvmlite

pip install llvmlite  # 未導入の場合

実行手順:

# ビルドLLVM_SYS_180_PREFIX不要
cargo build --release --features llvm

# 一括スモークテスト
tools/smokes/v2/run.sh --profile integration

# 個別スモークテスト
tools/smokes/v2/run.sh --profile integration --filter "<glob>"

# 単発実行
NYASH_LLVM_USE_HARNESS=1 $NYASH_BIN --backend llvm apps/tests/peek_expr_block.hako

# 有効化確認
$NYASH_BIN --version | rg -i 'features.*llvm'

💡 重要: 両方のラインのテストが通ることで、MIR14統一アーキテクチャの品質を保証

🔁 QuickでAST/LLVM系も実行したいとき

通常、quick は LLVM未ビルド時に AST/LLVM系テストを自動で SKIP します。 Quickでも実行したい場合は、先に LLVM 有効でビルドしてください:

LLVM_SYS_180_PREFIX=$(llvm-config-18 --prefix) cargo build --release --features llvm
tools/smokes/v2/run.sh --profile quick

テストランナーは LLVM 非対応時にヒントを出力しますbuildコマンドの案内

Notes

  • Using resolution: prefer nyash.toml aliases (SSOT). Some tests may enable NYASH_ALLOW_USING_FILE=1 internally for convenience.
  • Plugin warnings are informational; smokes are designed to pass without dynamic plugins.
  • Harness single-run may take longer due to link+exec; integration profile includes generous timeouts.
  • Binary selection in scripts: always use $NYASH_BIN (the runner sets it to target/release/hakorune when present, otherwise falls back to nyash).
  • Branding: uservisible messages prefer “Hakorune”; the legacy nyash binary remains for compatibility but may be gatedHAKO_ALLOW_NYASH=1)。

Dev Mode (defaults)

  • In v2 smokes, the quick profile exports NYASH_DEV=1 by default.
    • This enables CLI --dev-equivalent defaults inside Nyash:
      • AST using ON (SSOT + AST prelude merge)
      • Operator Boxes in observe mode (no adoption)
      • Minimal diagnostics; output parity is preserved
  • You can also run manually with nyash --dev script.hako.

Common commands

  • Quick suite (auto NYASH_DEV=1):
    • tools/smokes/v2/run.sh --profile quick
  • Focus JSON smokes:
    • tools/opbox-json.sh (Roundtrip/Nested, plugins disabled, generous timeout)
  • One-off program (VM):
    • $NYASH_BIN --backend vm --dev apps/APP/main.hako

Key env knobs

  • NYASH_DEV=1 — enable dev defaults (same effect as --dev).
  • SMOKES_DEFAULT_TIMEOUT — per test timeout seconds (default 15 for quick).
  • SMOKES_PLUGIN_MODE=dynamic|static — plugin mode for preflight (auto by default).
  • SMOKES_FORCE_CONFIG=rust_vm_dynamic|llvm_static — force backend config.
  • SMOKES_NOTIFY_TAIL — lines to show on failure tail (default 80).

Notes

  • Dev defaults are designed to be non-intrusive: tests remain behaviorcompatible.
  • To repro outside smokes, either pass --dev or export NYASH_DEV=1.

Heavy JSON probes

  • Heavy JSON tests (nested/roundtrip/query_min) run a tiny parser probe first.
  • The probe's stdout last non-empty line is trimmed and compared to ok.
  • If not ok, the test is SKIP (parser unavailable), not FAIL. This avoids false negatives due to environment noise or optional dependencies.