mmo-ref/tinymux
1
0
Fork 0
mirror of https://github.com/brazilofmux/tinymux synced 2026-08-13 00:23:11 -04:00
Code Issues Projects Releases Packages Wiki Activity
tinymux/docs/plan-lua-jit-bringup.md
Stephen Dennis 7472999178 test(lua/jit): make test-lua-jit runs full smoke with lua_jit 1 
Full smoke is green with mudconf.lua_jit forced on (1428/1428) and with
the default off.  Add an opt-in make target so CI/devs can exercise the
bytecode→HIR→DBT path without flipping production defaults.

Update plan-lua-jit-bringup and audit-coverage D4 for #1309 progress.
Engage assert (lua_run_ok) still waits on Hatsuhara #1317 jitstats keys.
2026-07-25 23:19:40 -06:00

3.2 KiB
Raw Permalink Blame History

Lua JIT bring-up plan (#1278 / #1309)

Owner: Grok (wip-grok) on #1309
PR: https://github.com/brazilofmux/tinymux/pull/1321
Related: #1278 (loader), #1316/#1317 (Hatsuhara: jitstats Lua counters), #1315 (Win64 tier-2 buffer)

Story (corrected)

  1. Lua interpreter works; smoke was green only on that path.
  2. lua_bc_load / read_int() treated Lua 5.4 varints as raw 4-byte ints since the pipeline landed (2026-03-18). Every compile failed → hir_lower_lua never ran.
  3. Fixing the loader revealed empty folds, wrong results, and a softcode- reachable hang in dbt_run (#1309). Not a regression of the loader — the code had never been exercised.

Phases

Phase Goal Status
0 Hygiene Claim #1309, worktree, plan done
1 Safety gate mudconf.lua_jit default off done (#1310)
2 Loader read_int varint done (#1310)
3a Correctness Trailing RETURN, MMBIN, code→memory, sref ATOI, nest done (PR #1321)
3b/c Suite Full smoke with lua_jit 1 done — 1428/1428
3d Opcodes Deeper HIR audit / HIR_NEG discrimination open
3e Limits Win64 tier-2 buffer (#1315 Hatsuhara) open
4 Default-on + engage assert Flip default; require lua_run_ok (#1317) next

Phase 3a root causes (fixed)

Symptom Cause
Empty return "hello" Trailing Lua RETURN overwrote real result with ""
Empty return 42 Fold ICONST→digit SCONST at return
Empty folds No hir_build_cfg → block_last=-1 → codegen skipped all insns
False multi_block RETURN marked next PC as leader
Skipped RETURN after ADD Extra pc++ past MMBIN
mux.args chaos / hang Map to CARGS; ATOI runtime_ref; copy rc.code into guest memory
Softcode JIT → fun_lua hang Refuse nested run_cached_program
Spin at guest PC 0 Zero code region (missing code copy) → illegal op → exit_with_pc(same)

Smoke verification (Unix/macOS, this tree)

# Default (interpreter-only Lua) — part of make test
make test   # or: cd testcases && ./tools/Smoke

# Force Lua JIT path — opt-in
make test-lua-jit
# equivalent: SMOKE_EXTRA_CONF='lua_jit 1' ./tools/Smoke

Results (2026-07-26):

Config Result
lua_jit 0 (default) 1428/1428
lua_jit 1 1428/1428

Coverage pin (before default-on)

Interpreter-only green is not enough. Need:

  1. Engage assert — after lua(), require lua_run_ok > 0 via jitstats()
    (Hatsuhara #1317 / #1316 — do not reimplement; merge then wire smoke TC).
  2. Optional control: delete HIR_NEG codegen → TC059/TC060 fail under lua_jit 1.

Claim map

Ticket / work Label
#1309 bring-up epic wip-grok
#1278 loader wip-kagura (landed via #1310)
#1316/#1317 jitstats Lua counters wip-hatsuhara
#1315 Win64 tier-2 buffer wip-hatsuhara

Next

  1. Land #1321; rebase if needed after #1317.
  2. Smoke TC: lua_jit 1 + jitstats() require lua_run_ok>0 (needs #1317).
  3. Windows: #1315 before flipping default on.
  4. Then Phase 4: mudconf.lua_jit default true.