Performance

Transpile speed

Measured on a game project (~180 source files, LuaJIT target) as of April 2026. Times vary by machine.

Scenario	Time
Initial build	~550ms
Incremental rebuild (watch mode)	~6-18ms

Initial build phase breakdown:

Phase	Time
Parse + bind	113ms
Type check	203ms
Transform	216ms
Print	10ms
Write	10ms

Watch mode architecture

In watch mode (tslua -p tsconfig.json --watch), incremental rebuilds are fast because:

• Incremental program update: typescript-go’s incremental.NewProgram diffs the changed file against its snapshot, avoiding a full reparse
• Async diagnostics: type-checking runs in a background goroutine after .lua files are written. Diagnostics arrive ~20-50ms later without blocking output
• Scoped semantic check: only the changed files are checked for import elision data, not the entire program
• fsnotify: file changes are detected via OS notifications, not polling

Phase breakdown (incremental clean edit)

Phase	Time
Program update	~1-3ms
Transform	~1-5ms
Print	~0.1-0.5ms
Write	~0.2-0.8ms
Build done	~6-18ms
Diagnostics (async)	+20-50ms

Runtime benchmarks

These benchmarks compare tstl (default) vs optimized emit mode as of April 2026. Optimized emit mode is early and only covers a few patterns so far (iterator allocation, tostring elision). The numbers below reflect what’s implemented today.

just bench              # run with LuaJIT (default)
just bench-lua          # show transpiled Lua output

LuaJIT

Benchmark	Time (tstl)	Time (opt)	Garbage (tstl)	Garbage (opt)
array_entries	0.238ms	0.028ms (8.5x)	988 KB	128 KB (-87%)
map_iterate	0.080ms	0.016ms (5.0x)	355 KB	96 KB (-73%)
set_iterate	0.014ms	0.011ms (1.3x)	65 KB	64 KB

Lua 5.1

Benchmark	Time (tstl)	Time (opt)	Garbage (tstl)	Garbage (opt)
array_entries	2.960ms	0.373ms (7.9x)	2600 KB	256 KB (-90%)
map_iterate	0.974ms	0.463ms (2.1x)	897 KB	193 KB (-78%)
set_iterate	0.807ms	0.389ms (2.1x)	551 KB	129 KB (-77%)

Other benchmarks (array_iterate, array_push, string_iterate, string_concat) show no significant difference between modes. Run the full suite with just bench.

The wins come from iterator optimizations in optimized emit mode. Map/Set for-of loops use custom stateless Lua iterators that walk the internal linked list directly, avoiding per-step closure and table allocations. See Emit Mode for details.