[esrrhs]

Summary

This PR speeds up interpreter calls in two related ways, landed together:

1. Direct INTERP call fast paths. Exact-arity INTERP-to-INTERP named and closure calls go through InterpreterExecute / InterpExecuteInContext without copying VmFunction or entering GuardJitEntry / an extra LuaLocScope. Vm::FindFunction(std::string_view) returns a pointer into the function table. CALLNAME / CallByNameArgs take string_view and no longer heap-allocate std::string on the success path (allocation remains on error paths).

2. Looped tail calls. return f(...) (not parenthesized return (f(...))) emits TAIL / TAILNAME. The common exact-arity, non-vararg, no-box self-tail reuses the current frame in place (inst.c = kTailSelfExact, no name lookup). Mutual recursion, vararg, Multi, JIT/native callees, and no-box→boxed or root-frame growth use a generic fallback and a TLS trampoline so C-stack depth stays one. Normal CALL / CALLNAME do not pay a per-tail setup cost.

Semantics / tests

New fixtures and cases cover self, mutual, vararg, upvalue, multi-return, deep INTERP tails (n=100000), and tail error locations (TailErr.boom line 6, call_missing line 13, including deep INTERP n=20000). File-level functions cannot be reassigned, which is what allows self-TAILNAME to skip the name lookup.

Locally, 562 related unit tests (test_basic, closure, runtime, vm_cvar_call, test_native, algo, jitter, exception) passed under an addr2line bwrap stub used only to keep Boost.Stacktrace throw-heavy tests from hanging on symbolization. This is not a claim that the full suite or CI is green.

Measurements

Comparable Release, same harness, debug_mode=false, min_time=0.3s, 3 sequential reps, median Time ns/call. Three libraries (master at 49c63c7, call-opt alone, and this combined change) via LD_LIBRARY_PATH. No concurrent compile during the run. INTERP TailRec at large n ran in a forked child so C-stack overflow is recorded as a crash rather than killing the driver.

master → this PR (combined call-opt + tail):

• EmptyCall n=10000 INTERP: 829704 → 559517 ns (1.48×)
• non-tail Recursion (fib) n=20 INTERP: 1888952 → 1298006 ns (1.45×)

Tradeoff versus call-opt alone (not an unqualified all-path speedup): the combined change is about 4% slower on EmptyCall and about 11% slower on non-tail Recursion than the call-opt-only intermediate (EmptyCall 538246 → 559517; Recursion 1165972 → 1298006). The residual is likely interpreter loop code layout. There is no new public opt-in flag; tail is on by default because of the tail-call wins and stack safety below.

TailRec INTERP (self tail):

• n=200: call-opt 11457 → 4777 ns (2.4×)
• n=1000: call-opt 64729 → 23477 ns (2.76×)
• n=100000: 2449587 ns (~2.45 ms) on this change; master and call-opt overflow the C stack at n=5000.

Notes

Default CompileConfig / debug_mode and unrelated caches are unchanged.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 87.83270% with 32 lines in your changes missing coverage. Please review.
✅ Project coverage is 89.73%. Comparing base (49c63c7) to head (ab7a7e8).

Files with missing lines	Patch %	Lines
src/interp/interpreter.cpp	85.90%	31 Missing ⚠️
src/interp/codegen.cpp	96.00%	1 Missing ⚠️
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #55      +/-   ##
==========================================
- Coverage   89.77%   89.73%   -0.04%     
==========================================
  Files         125      125              
  Lines       27317    27554     +237     
==========================================
+ Hits        24523    24725     +202     
- Misses       2794     2829      +35     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.