Behavior Differences: luars vs C Lua 5.5

May 19, 2026 · View on GitHub

This document records all known behavioral differences between luars (Rust implementation) and the official C Lua 5.5 reference interpreter.

1. No C API / testC support

luars is a pure Rust implementation with no C API (lua_State*, lua_pushinteger, etc.). The official test suite's T (testC) library is unavailable, causing the following test files/sections to be fully or partially skipped:

Test file	Scope	Reason
`api.lua`	whole file	C API tests
`memerr.lua`	whole file	C memory-error injection
`code.lua`	whole file	opcode-level tests via `T.listcode`
`main.lua`	whole file	interactive interpreter / CLI arg tests
`db.lua`	whole file	full debug-library tests (hook-dependent)
`coroutine.lua`	2 sections (~L595, ~L1062)	C API coroutine tests
`events.lua`	1 section (~L331)	userdata event tests
`errors.lua`	1 section (~L94)	C function error-message tests
`gc.lua`	1 section (~L401)	userdata GC tests
`strings.lua`	2 sections (~L474, ~L534)	`pushfstring` / external-string tests
`nextvar.lua`	2 sections (~L146, ~L665)	table-size / table-lib-on-non-tables tests

2. Rust host API string view

At the Lua level, luars now stores strings as raw byte strings, so arbitrary bytes are preserved just like in C Lua.

The remaining difference is on the Rust host side:

LuaValue::as_str() only returns Some(&str) when the underlying bytes are valid UTF-8.
LuaValue::as_bytes() should be used when exact byte preservation matters.
LuaState::create_bytes() can be used to create Lua strings from arbitrary bytes without forcing UTF-8.

This is an embedding/API difference, not a Lua-language semantic difference.

3. No locale-dependent number parsing

Number parsing always uses . as the decimal separator and is unaffected by os.setlocale.

Affected: literals.lua (~L300–L325) — locale test block skipped via if false then.

4. No C module loading

package.loadlib always returns "loadlib not implemented".
The package.cpath searcher (searcher 3) always fails; .so / .dll modules cannot be loaded.

Affected: attrib.lua (~L265, ~L335) — C library / external-string tests skipped.

5. debug library limitations

The debug library is partially implemented. debug.setuservalue and debug.getuservalue are not implemented. Because we are different userdata implementations.

6. Bytecode format

luars uses its own bytecode format, incompatible with C Lua binary chunks. string.dump output can only be loaded by luars itself. Upvalue names are not included in the dump.

The on-disk format still distinguishes UTF-8 string constants from raw byte-string constants so luars can round-trip non-UTF-8 data through string.dump and load, even though the runtime no longer has a separate binary value tag.

Affected: calls.lua (~L484) — binary-chunk header tests skipped.

7. GC implementation differences

Stack shrinking during GC is not implemented.

8. Coroutine limitations

Multi-value .. (OP_CONCAT) across a yield does not resume correctly — missing finishOp continuation for CONCAT. Test in coroutine.lua (~L982) skipped.
Stack-overflow limit differs from C Lua's 1 MiB policy. Test in coroutine.lua (~L839) skipped.

9. `os.date` DST flag

os.date does not compute the isdst (daylight saving time) field correctly; it defaults to false.

10. Long-string address reuse

C Lua may reuse the same address for identical long-string constants (useful with const tags). luars allocates separate objects.

Affected: literals.lua (~L213) — long-string reuse test skipped.