[Feature request] Add a Lua 5.1 polyfill for __len and rawlen

[SkyyySi]

The length-operator isn't overloadable in PUC Lua 5.1 at all, and can only be overloaded in LuaJIT when using a build with the compilation-flag -DLUAJIT_ENABLE_LUA52COMPAT enabled (which isn't the case for basically all pre-built distributions).

I would appreciate it if a polyfill were to be automatically inlined when targeting Lua 5.1. A basic version could look like this:

global rawlen = (object) ->
    #object

_len_0 = (object) ->
    if metatable := getmetatable(object)
        if len_meta := rawget(metatable, "__len")
            return len_meta(object)

    #object

It could then be inserted whenever getting the length of an object, i.e.:

foo = ["x", "y", "z"]
bar = #foo

would compile to

rawlen = function(object)
    return #object
end
local _len_0
_len_0 = function(object)
    -- ...
end
local foo = {
    "x",
    "y",
    "z"
}
local bar = _len_0(foo)

---

The code above does work as-is, but I would personally not recommend using that. Instead, the code below is more robust / "correct", since it handles some additional edge cases.

--- Check whether this polyfill is actually needed before inserting it. It is
--- only required when using Lua 5.1, as well as not using LuaJIT with Lua 5.2
--- compatibility mode enabled. This is mainly here to avoid doing unnecessary
--- computations when creating a cross-version script.
const _len_0 = if (_G._VERSION == "Lua 5.1") and (_G.rawlen == nil)
    import type, error, rawget, select, debug from _G

    global rawlen = (object) ->
        #object

    const _getmetatable_0 = do
        const gloabl_getmetatable = _G.getmetatable

        --- The regular / global `getmetatable()`-function can be overloaded
        --- using the `__metatable`-field of a metatable. This is basically
        --- never used in practice (because why on earth would you do this?),
        --- but for the sake of correctness, this wrapper at least prevents
        --- issues with `__metatable` being set to an unexpected type.
        const wrapper = (object) ->
            const metatable = gloabl_getmetatable(object)

            if type(metatable) == "table"
                metatable
            else
                nil

        --- If possible, `debug.getmetatable()` is used instead of the
        --- global `getmetatable()`-function. This is because the latter
        --- could get tricked by something like this:
        ---
        --- ```yuescript
        --- tb = {
        ---     <metatable>: {
        ---         <len>: () => 5
        ---     }
        ---     <len>: () => 3
        --- }
        --- 
        --- --- This may incorrectly print 5 instead of 3 or 0
        --- print(#getmetatable(tb))
        --- ```
        ---
        --- Note that some environments disable access to the `debug`-library.
        if (type(debug) == "table") and (type(debug.getmetatable) == "function")
            debug.getmetatable
        else
            wrapper

    const get_function_name = if (type(debug) == "table") and (type(debug.getinfo) == "function")
        import getinfo from debug
        () -> getinfo(2, "n").name
    else
        () -> "_len_0"

    (...) ->
        do
            const argc = select("#", ...)
            if argc != 1
                const function_name = get_function_name()
                error("function %s() expected exactly one argument, got: %d"::format(
                    function_name
                    argc
                ))

        const object = ...

        do
            --- Throw an error if the type of `object` doesn't support getting
            --- it's length.
            const type_of_object = type(object)
            if type_of_object not in ["string", "table", "userdata"]
                --- This is the same error message that Lua would throw when
                --- doing something like this:
                ---
                --- ```lua
                --- print(#(false))
                --- ```
                error("attempt to get length of a %s value"::format(
                    type_of_object
                ))

        const metatable = _getmetatable_0(object)

        --- `object` has no metatable -> return its raw length.
        if metatable == nil
            return #object

        --- Use `rawget()` to prevent being tricked by `__index`, e.g.:
        ---
        --- ```yuescript
        --- tb = {
        ---     <>: {
        ---         <index>: {
        ---             __len: () => 7
        ---         }
        ---     }
        --- }
        ---
        --- --- This will incorrectly print 7 instead of throwing an error
        --- print(getmetatable(tb).__len(tb))
        --- ```
        const len_meta = rawget(metatable, "__len")

        --- `object` has no `__len`-metamethod -> return its raw length.
        --- Note: `__len` might also be a callable table or userdata-object
        --- instead of a function, which is why no strict type check is
        --- performed here.
        if len_meta == nil
            return #object

        len_meta(object)
else
    (object) ->
        #object

And here's a bunch of tests:

assert(_len_0({
    "foo"
    "bar"
    "biz"
    "baz"

    n: 3

    <len>: () => @n
}) == 3)

assert(_len_0({
    <>: {
        <index>: {
            __len: () => 7
        }
    }
}) == 0)

assert(_len_0({
    <>: {
        <metatable>: {
            __len: () => 7
        }
    }
}) == 0)

assert(_len_0({
    <metatable>: {
        <len>: () => 5
    }
    <len>: () => 3
}) == 3)

assert(_len_0("test") == 4)

assert(not pcall(_len_0, 123))

assert(not pcall(_len_0, ->))

[pigpigyyy]

Thanks for the suggestion! What you're proposing would essentially involve adding a runtime library for YueScript. However, I'm recently considering implementing a feature that would generate the runtime module separately, rather than embedding the library code in every individual file. This approach might offer some advantages in terms of maintainability and efficiency.

[gphg]

Can we just have jit as another target? It is based on 5.1, with some features on 5.2 and 5.3.