perf(lib): streamline print function using inline caching and table.c…

[Stricky4]

…oncat

Optimized the print function by streamlining the argument processing flow.

• Initialized buffer and cached global functions (tostring, table.concat) in a single assignment.
• Used a direct index-based loop for type conversion, avoiding repeated table insertions.
• Benchmarks show significant performance gains, reducing execution time by up to 51%.
• Maintained full compatibility with nil values and varied data types.

[Cheatoid]

The file has mixed indentation tabs/whitespace. Perhaps fix that too. (use tabs for smaller file size)

[Stricky4]

thx

[Cheatoid]

Mixed quote style. select uses " so stick with that.

[Stricky4]

thx

[Cheatoid]

{...} creates a new table every call which is bad, it better be done in C.
I would like to see benchmark against my raw table.pack version:

-- Localized global functions at module-level for better performance
local rawget, rawset, tostring, table_concat, table_pack = rawget, rawset, tostring, table.concat, table.pack

print = function(...)
    local packed = table_pack(...)
    for i = 1, packed.n do
        rawset(packed, i, tostring(rawget(packed, i)))
    end
    return Console.Log(table_concat(packed, "\t"))
end

[JbLaDeb]

I ran multiple benchmarks, and table.pack is consistently slower:

Run 1:
v1 ({...}) : 2.69s
v2 (table.pack) : 3.52s (~30% slower)

Run 2:
v1 ({...}) : 1.68s
v2 (table.pack) : 3.14s (~86% slower)

Both implementations allocate a table anyway, so table.pack doesn’t avoid allocations. In practice it introduces extra overhead (the .n field) and seems to be less JIT-friendly

Given that, {...} is not only simpler but also significantly faster here, so we’d prefer keeping the current implementation

[Cheatoid]

nanos world does not use LuaJIT, it uses Lua 5.4.8. Did you benchmark this ingame? Also, yes they both allocate, but table.pack does it in C.
btw, just in time I released benchmark library, check it out:
https://github.com/Cheatoid/nanos-world-vault/tree/main/library#benchmark

[JbLaDeb]

Yeah, thanks for the correction regarding Lua 5.4, that said, it doesn’t change the conclusion since the benchmarks still show {...} is consistently faster

[vugi99]

I have a better solution :

local function print_voltaism(...)
    return Console.Log(string.format(string.rep("%s\t", select("#", ...)), ...))
end

My testing code

local args = {"test", "args", nil, 2, {}, true, 2.5, n=7}
local count = 100000

local ret_current = NanosUtils.Benchmark("print_current", count, print_current, table.unpack(args, 1, args.n))
local ret_stricky4 = NanosUtils.Benchmark("print_stricky4", count, print_stricky4, table.unpack(args, 1, args.n))
local ret_cheatoid = NanosUtils.Benchmark("print_cheatoid", count, print_cheatoid, table.unpack(args, 1, args.n))
local ret_voltaism = NanosUtils.Benchmark("print_voltaism", count, print_voltaism, table.unpack(args, 1, args.n))

print("print_current", tostring(ret_current) .. "ms")
print("print_stricky4", tostring(ret_stricky4) .. "ms")
print("print_cheatoid", tostring(ret_cheatoid) .. "ms")
print("print_voltaism", tostring(ret_voltaism) .. "ms")

Results :

[Cheatoid]

Maybe maybe... add printf for convenience (after print is defined):

do
    local _print, string_format = print, string.format
    --- Prints a formatted string to the console.
    ---@param ... any # Format string and values to format.
    printf = function(...) return _print(string_format(...)) end
end

( would deprecate this )