Lua C++ Classroom

The Roblox engine is written in a combination of C++ and Lua, with the code that performs computationally intensive operations written in optimized C++, while game logic and scripts are written in Lua, for ease of development. For this model to be effective, it requires that the transitions between Lua and C++ be as fast as possible, as any time spent in this no man’s land is essentially just wasted milliseconds.

Lua's API is not designed to register C classes to Lua, only C functions that have the signature int(luaState.), that is, a function that takes an luaStatepointer as an argument and returns an integer. Actually, that is the only C data type that Lua supports in registering. Many Lua developers might find: for method calls more consistent with the vast majority of object-oriented Lua code. Advantages of closure-based: Closure-based instances can have truly private fields, so that the users of your class cannot accidentally or intentionally get to them. Access to private fields is faster with closure-based classes. Using Lua with existing projects written in C/C. The Lua language provides a set of API functions for communication between C/C and the Lua programming language. You can use these functions in your C code after you include the lua.h header file or lua.hpp for C source code. You can rely on this set of functions, but sooner or later you'll. Updated There is an undocumented newproxy feature in Lua 5.1 that allows implementation of finalizers on tables; Lua 5.2 removed that feature as it added support for gc metamethod on tables. No yielding between Lua and C code: coroutine.yield call across Lua/C boundary fails with attempt to yield across metamethod/C-call boundary. Coding Lessons for Kids in Buffalo, NY. Coding takes practice, and we believe getting kids to code consistently throughout the year is the only way they'll really learn to code. TheCoderSchool has a staff of over 700 Code Coaches all around the country, so we know a thing or two about mentoring students in computer science.

Over the past couple of months, we’ve been rolling out various improvements to this part of the system. One part specifically—the actual invocation of C++ methods from Lua—was particularly interesting, as it led to considerable speed improvements and required digging around in the guts of Lua to understand how things worked under the hood.

Lua Objects

We ended up modifying the Lua VM itself, but before we get to that, we need to lay some groundwork.

Compilers, VM, and bytecode

When Lua source code is compiled, it’s compiled into Lua bytecode, which the Lua VM then runs. Lua bytecode has around 35 instructions in total, for things like reading/writing tables, calling functions, performing binary operations, jumps and conditionals, and so on. The Lua VM is register-based, as opposed to being stack-based like many other VMs, so part of what the compiler does when it generates bytecode is determine which registers each instruction should use.

Each instruction has the form “OP_CODE A B,” or “OP_CODE A B C,” where “OP_CODE” is the opcode (for example, CALL for calling a function) and A/B/C are the opcode arguments. The arguments (or registers) aren’t actual values. Instead, they are indices that point into one of two tables: the constant table (written Kst(..)) or the register table (written R(..)).

Interlude

For a detailed description of Lua bytecode, see “A No-Frills introduction to Lua 5.1 VM Instructions.” It’s a lot more exciting than it sounds; I promise!

To give you a feel for what Lua bytecode looks like, we’re going to go over some simple programs first and then progress to some more relevant examples.

Interlude

Using the Chunkspy utility, we can disassemble Lua bytecode into Lua assembly and get a listing of the code, as well as the constant table, so we can essentially see what bytecode gets generated for any given Lua source code.

Basic bytecode examples

A simple program like “x = 10” compiles into:

The first two lines show the constant table (with the string value “x” in slot 0 and the integer value 10 in slot 1), and the following two lines are the disassembled opcodes.

[Line 1] Looking up the LOADK opcode in “No Frills,” we see that it has the form “LOADK A Bx — R(A) := Kst(Bx).” So, LOADK has two arguments (registers A and B) and its operation is to assign the value found in the constant table at the slot given by the second register, Kst(Bx), to the register table in the slot given by the first argument, R(A). “Bx” just means that because the opcode only has two arguments, the B register is extended and assigned more bits.

[Line 2] SETGLOBAL has the form “SETGLOBAL A Bx — Gbl[Kst(Bx)] := R(A).” It assigns a value to the global table using the key given by the constant table at the slot of the second argument. Because the second argument is 0 and the value of the constant table at 0 is “x,” it writes something to the global table using the key “x.” Whatever is in the register table at the slot given by the first argument is what’s being written, which the previous instruction loaded with the value 10.

Let’s look at a slightly more complicated example, “x = 10; y = x.” I’ll leave the manual execution of the code as an exercise for the reader. 🙂

Bytecode for function calls

Let’s look at the code generated for “foo(10):”

To execute function calls, the function must be loaded into the first register and arguments into subsequent ones. The semantics for “CALL A B C” are such that A holds the function, B is the number of arguments (actually, it’s the number of arguments +1, due to the way “…” is implemented), and C is the number of return values (again, it’s the number of return values +1, to handle multiple return values).

We’re familiar with the first two lines; they load a value into register table slot 0 and the value 10 into register table slot 1. The third line is what performs the function call, using the value in register A (register table slot 0, which was loaded with “foo”), with B specifying the number of arguments, and C the number of return values (remember, both B and C should have 1 added to their values). Before the function can be called, the VM also verifies that the value in R(A) is in fact callable.

Interlude — Metafunctions, metatables, and userdata

Lua has a mechanism that allows users to extend the functionality of tables by associating a metatable with an existing table. The metatable contains fallback methods that are invoked if a certain method or operation can’t be performed on the main table (see https://www.lua.org/pil/13.html for a thorough description).

For our purposes, the most relevant entries in the metatable are the “__index” and “__call” fields. __index is used when looking up an element in a table, so the code “local x = my_table[10]” would first call the __index method on my_table. If that failed, it would instead try to call __index on my_table’s metatable. __call is similarly used when you try to treat something as a function and call it “local x = foo(42),” for example

In order for Lua and C++ to interoperate, they need some way to be able to share functions and data. Lua facilitates this by providing a data type called UserData. UserData objects can be created in C++ land, and because they are native Lua data types, they can be adorned with metatables that allows Lua code to interact with them as if they were just regular Lua objects.

Member function calls

Okay, back to looking at some bytecode! This next example is a bit more interesting because it shows what happens when you have code like “foo:bar(10),” which is calling the bar method on the instance foo (an instance of the class Foo).

The new thing here is the self instruction [line 2], which we haven’t seen before. Self has the syntax “SELF A B C — R(A) := R(B)[RK(C)]; R(A+1) := R(B),” so let’s break this down. In the register table at slot R(A), it will place the result of looking up the table in register slot R(B) using the key in slot RK(C). It will also copy whatever was in slot R(B) into slot R(A+1), but more