hakorune

Files

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

configs

builder/vm: stabilize json_lint_vm under unified calls

2025-09-28 12:19:49 +09:00

lib

Phase 21.7 normalization: optimization pre-work + bench harness expansion

2025-11-13 16:40:58 +09:00

profiles

fix(perf): stabilize MIR emit for ny-llvmc/EXE benchmarks

2025-11-13 17:23:48 +09:00

README_CLAUDE_REMINDER.md

phase: 20.49 COMPLETE; 20.50 Flow+String minimal reps; 20.51 selfhost v0/v1 minimal (Option A/B); hv1-inline binop/unop/copy; docs + run_all + CURRENT_TASK -> 21.0

2025-11-06 15:41:52 +09:00

README.md

Phase 21.7 normalization: optimization pre-work + bench harness expansion

2025-11-13 16:40:58 +09:00

run_quick.sh

feat(phase21.5/22.1): MirBuilder JsonFrag refactor + FileBox ring-1 + registry tests

2025-11-10 19:42:42 +09:00

run.sh

feat(phase21.5): MirBuilder optimization prep + crate EXE infrastructure

2025-11-11 02:07:12 +09:00

README.md

Smokes v2 — Minimal Runner and Policy

Policy

Use [SKIP:] prefix for environment/host dependent skips.
- Examples: [SKIP] hakorune not built, [SKIP:env] plugin path missing
- Keep reasons short and stable to allow grep-based canaries.
Prefer JSON-only output in CI: set NYASH_JSON_ONLY=1 to avoid noisy logs.
Diagnostics lines like [provider/select:*] are filtered by default in lib/test_runner.sh.
- Toggle: set HAKO_SILENT_TAGS=0 to disable filtering and show raw logs. HAKO_SHOW_CALL_LOGS=1 also bypasses filtering.

Helpers

tools/smokes/v2/lib/mir_canary.sh provides:
- extract_mir_from_output — between [MIR_BEGIN]/[MIR_END]
- assert_has_tokens, assert_skip_tag, assert_order, assert_token_count
tools/lib/canary.sh provides minimal, harness-agnostic aliases:
- extract_mir_between_tags — same as extract_mir_from_output
- require_tokens token... — fail if any token missing

Notes

Avoid running heavy integration smokes in CI by default. Use --profile quick.
When a test depends on external tools (e.g., LLVM), prefer [SKIP:<reason>] over failure.

Quick tips

EXE-heavy cases (e.g., phase2100/*) may take longer. When running quick with these tests, pass a larger timeout like --timeout 120.
Smokes v2 auto-cleans temporary crate EXE objects created under /tmp (pattern: ny_crate_backend_exe_*.o) after the run.