Consolidate API object code

[TurkeyMcMac]

This PR reduces repetition in the code that implements userdata objects such as VoxelManip and Settings. Overall, it removes over 600 lines of code. It should also decrease the size of the executable, but I haven't measured that.

To do

This PR is Ready for Review.

How to test

There are many changes in the PR, and it's possible that some are erroneous. Someone should look through them to make sure this is not the case.

[sfan5]

Hmm, not sure if I like the closure/upvalue solution even if it is a bit faster. But we'll get to this when the other PR is done.

[TurkeyMcMac]

I probably won't rebase this unless the concept is approved.

[TurkeyMcMac]

Nevermind, this rebase at least was easy.

[sfan5]

Okay so like I hinted at I'm not a fan of the extra complexity of the upvalue/closure solution and I think we should keep using luaL_checkudata.
In that case what should be done to reduce duplication is to turn checkobject into a templated helper and the same registerClass change as already done now.

[TurkeyMcMac]

I have changed checkObject to be templated. That does clean up the code a bit. I have not removed methodCheckObject at the moment.

I should do a benchmark using PUC Lua where a RNG is used in a loop. This is a realistic situation to benchmark method overhead.

[TurkeyMcMac]

Realistic benchmark using PUC Lua:

local rng = PcgRandom(123)
local results = {}
local before = core.get_us_time()
for i = 1, 10000 do
	results[i] = rng:next()
end
local after = core.get_us_time()
core.debug("Benchmark:", after - before)

Time with methodCheckObject: ~1200
Time with checkObject: ~1600

[Desour]

All that luaL_checkudata does different is where it gets the class's userdata from:

minetest/lib/lua/src/lauxlib.c

Lines 124 to 137 in b123305

	LUALIB_API void *luaL_checkudata (lua_State *L, int ud, const char *tname) {
	void *p = lua_touserdata(L, ud);
	if (p != NULL) { /* value is a userdata? */
	if (lua_getmetatable(L, ud)) { /* does it have a metatable? */
	lua_getfield(L, LUA_REGISTRYINDEX, tname); /* get correct metatable */
	if (lua_rawequal(L, -1, -2)) { /* does it have the correct mt? */
	lua_pop(L, 2); /* remove both metatables */
	return p;
	}
	}
	}
	luaL_typerror(L, ud, tname); /* else error */
	return NULL; /* to avoid warnings */
	}

You could try to store the userdatas in an array table in the registry instead of using a string index.

[TurkeyMcMac]

Good idea.

[TurkeyMcMac]

With LuaJIT, the integer index solution seemed like a lot of complexity for no performance gain, perhaps even a small performance loss. I decided to remove the optimizations from this PR altogether. They can be added separately.

[sfan5]

This really should not be micro-optimized, with the numbers from the earlier benchmark the idiomatic solution with luaL_checkudata is 40 nanoseconds slower than upvalues.

[TurkeyMcMac]

Probably true. In any case, optimization is no longer a goal of this PR.

[sfan5]

LGTM
if you've tested it reasonably feel free to merge