Add fnRegsFromStruct for comptime-generation of Lua function registrations

[nurpax]

No description provided.

[nurpax]

CI failed.. Forced pushed to update.

Imagine if both of these paths worked on all Lua versions lol

    if (zlua.lang == .lua51 or zlua.lang == .luau or zlua.lang == .luajit) {
        lua.newTable();
        lua.registerFns("fnregs", funcs);
    } else {
        lua.newLib(zlua.fnRegsFromStruct(MyLib));
        lua.setGlobal("fnregs");
    }

[robbielyman]

love this!! really excited to add it :)

[nurpax]

Hi! Is moving the comptime keyword to caller site the more idiomatic way in Zig? I guess I'm used to doing it the way I did it in the original version.

For example, I see comptime used in the library, not at call time, like in the initCompTime function here:

https://ziglang.org/documentation/master/std/#src/std/static_string_map.zig

I don't have a super strong opinion about, but I guess I have a preference for version that doesn't require the callsite to mark the call comptime.

BTW, I intentionally had this working only for structs because I didn't think this sort of "get decls from a namespace" makes so much sense e.g. in enums or unions. Or it may at least be that some public funcs in an enum likely aren't zlua wrappable. In this case, requiring a struct and giving an error may guide the user of this library to the right path. OTOH, if you had some cool uses in mind, then being overly restrictive is not that great idea either!

[robbielyman]

no, i think what you did was right! the reason for comptime here is actually to move it to a testable part of the API contract. sorry for also unrelatedly fiddling with the implementation a little—I liked proving to myself that my code was equivalent to yours even though it doesn't have an explicitly comptime block.

[nurpax]

Would you be willing to accept a PR that moves comptime back to ziglua so that users of this API don't need to explicitly add comptime? That seems to be the way it's usually done in std library at least.

[robbielyman]

I would, if you insist, but I claim that you are misunderstanding what I am saying.

• the use of the comptime keyword in the caller here makes no difference and does not need to be done in consumer code.
• the purpose of the comptime keyword in the test is so that in the future if the function body is altered to include runtime code, CI will fail
• the code in the body of the function is executed entirely at compile time currently, even though there is no explicit comptime block. please read the function carefully: the main body of work that it does constructs a []const FnReg by modifying a comptime var. All of the code up to const final = funcs; therefore must be run at compile time.

[nurpax]

Oh nice! Thanks for clarifying and sorry for the confusion. I had indeed misunderstood what you said before. In this case there's no need to follow up with a PR. :)