Mission Scripting (Overview)

This article may need to be rewritten. Please help improve this article. The discussion page may contain suggestions.

This article deals with the general overview on the mission scripting in GTA 3D series.

Introduction

The original Mission Script looked something like this (taken from Vice City debug.sc file):

IF IS_BUTTON_PRESSED PAD2 RIGHTSHOULDER1
AND flag_create_car = 1
AND button_press_flag = 0
	IF IS_CAR_DEAD magic_car
		DELETE_CAR magic_car
	ELSE
		IF NOT IS_PLAYER_IN_CAR player magic_car
			DELETE_CAR magic_car
		ELSE
			MARK_CAR_AS_NO_LONGER_NEEDED magic_car
		ENDIF
	ENDIF 
	flag_create_car = 0
	initial_car_selected = 0
	button_press_flag = 1
ENDIF

Easy to read and understand, it is fairly basic so anyone with an idea of basic coding (or maybe even English) can understand it. However, very little code came with the game like that. The majority of the mission script comes in a file called main.scm (although in San Andreas there are alternate mains and external scripts, but they all follow the same basic format - hex codes). Example, for the code:

IF IS_CAR_DEAD magic_car
	DELETE_CAR magic_car

The equivalent in the main.scm would look something like this:

D6 00 04 00 19 01 02 45 0E 4D 00 01 FE 3D 87 02 A6 00 02 45 0E

Script instructions

The following information is only valid for GTA 3, Vice City and San Andreas. GTA 4 has completely different scripting engine. For more details on it see the SCO format article.

A scm file itself is a bytecode containing instructions telling to the game what to perform. An instruction consist of an opcode and its parameters (it there are any). Sometime the whole script instruction is called opcode.

Opcode

Each instruction is represented by a number – opcode (operation code), which is UINT16. By this number the game engine identifies an action to do. Say, opcode 0001 tells to wait for amount of time, 0003 shakes the camera, 0053 creates a player, etc.

This is how an opcode 0001 looks in a scm file:

0100 04 00

• First part is the opcode number in a little-endian format.
• Second part is the data type
• Third part is a parameter value

When a mission script is disassembled, opcodes are written in a human-readable format. The example above will look something like this:

wait 0

This is made for the end-user convenience only. The game does not know what the word wait means, but it knows what the opcode 0001 is, so when a mission script is assembled the commands are written back in raw byte form.

As it has been said, an opcode is UINT16 number. It means the minimum opcode is 0000 and maximum opcode is 0xFFFF. However due to a specific of the SCM language, any numbers above 0x7FFF denote negative conditional opcodes. More on this read there. The original unmodded game supports a way smaller amount of opcodes (0A4E for San Andreas), but there are tools adding new ones, most notably CLEO library.

After an opcode number the data types and parameter values follow^[*].

Data types

Data types is a single byte written before any parameter^[*]. The purpose of it is to tell to the game how much bytes to read and what kind of data it is. As it's seen from the table two bytes 02 00 could have 3 different meanings as a parameter: if it's preceeded by a data type of 2 it is a global variable ($2), data type of 3 – local variable (2@), or 16-bit number (2), so without data type the game wouldn't determine the parameter type.

Data type	Size (bytes)	Game	Description
00	0		end of argument list (004F, 0913 and similar)[*]
01	4		immediate 32 bit signed int
02	2		global int/float var
03	2		local int/float var
04	1		immediate 8 bit signed int
05	2		immediate 16 bit signed int
06	4		immediate 32 bit float
07	6		global int/float var array[*]
08	6		local int/float var array[*]
09	8		immediate 8 byte string
10	2		global 8 byte string var
11	2		local 8 byte string var
12	6		global 8 byte string var array[*]
13	6		local 8 byte string var array[*]
14	1+x		immediate varlen string - first you read 1 byte which gives you length of the rest which is text
15	16		immediate 16 byte string
16	2		global varlen string var
17	2		local varlen string var
18	6		global varlen string var array[*]
19	6		local varlen string var array[*]
>31	8		immediate 8 byte string[*]

^ Arrays are assembled as 2 UINT16s and 2 bytes:

2b - UINT16 - array offset[*]
2b - UINT16 - array index[*]
1b - BYTE   - flags[*]
1b - BYTE   - array size

^ An array offset basically is a variable number. If it's a global array, the offset is a global variable index from which the array begins. For example, if the global array offset is 150 (96 00) it means that the first element of the array is $150, the second one is $151, etc. Same valid for the local arrays (offsets are local variable indexes).

^ An array index is a variable number (global or local one) that holds the value of array index. For example, if array index is 3 (03 00), the game will read either global variable $3 or local variable 3@ depending on the flags. This variable holds the number which is array element ID to work with. For example, if the array index is $3, and $3 holds number 5, the game will read 5th element of the array.

^ Flags

0x0 - integer array with local variable as index	0x80 - integer array with global variable as index
0x1 - float array with local variable as index	0x81 - float array with global variable as index
0x2 - short string array with local variable as index	0x82 - short string array with global variable as index
0x3 - long string array with local variable as index	0x83 - long string array with global variable as index

So, 07 96 00 03 00 01 05 will be interpreted as $150(3@,5f).

Parameters

The game engine knows amount of parameters for each opcode (1 for 0001, 2 for 0004, 13 for 014B, etc). If the script contains another number of parameter it causes a crash.

The parameters could be one of following kinds:

• Immediate values
  • integer numbers
  • floating-point numbers
  • short strings (up to 7 characters and null-terminator byte)
  • long strings (up to 15 characters and length byte)
• Variables
  • global variables
  • local variables
• Arrays

Notes

^ In GTA 3 and Vice City short strings (8 bytes) have no data type preceeding it. If the byte does not fit data type range (0..6 for GTA 3 and VC), it's recognized as a beginning of a string and next 8 bytes are read.

^ Some opcodes have variable amount of parameters. Most known opcode is 004F that creates a new thread and passes arguments to it. The number of such parameters could vary, so the special data type denotes the end of parameters.

The maximum amount of parameters for any opcode is 16 for GTA 3 and GTA VC, 32 for San Andreas.

^ Opcode 05B6 is a special opcode that defines a table. Immediately after opcode number the stream of data (128 bytes) follow, without a data type.

This section is incomplete. You can help by fixing and expanding it.

Cracking the SCM

As has been said, very little of the code was supplied with the game in a decompiled state (only two small files, both test scripts), so how, as asked, do we create our own scripts based on the original? With a decompiler - but how do these work (no decompilers have been provided by Rockstar).

The original SCM format was cracked shortly after the release of GTA 3 (the first game to use this mission coding method), with people having to first figure out what all the sections did (there are 5 segments is an SCM - memory, objects, mission defines, MAIN and missions (GTA SA has more, but only one of these (global variables) has had its use determined), where they started/ended etc, figuring out how many parameters each OpCode had and a lot more. Once this was done, they knew where each OpCode began and ended, so they could split them up to make it more readable, but the data on what each one does was lost in the compiling, so they still only had something that looked like this:

:label035F78
0001: 0?
00D6: 0?
0256: 4??

That doesn't still doesn't mean a lot though, so people had to try figure out what the different OpCodes meant.

(Note: this code is in early Mission Builder format:

:labelxxxxxx means this code was originally at this offset in the mission script (the 'label' is added in by the decompiler)
x? means a one byte number
x?? means a variable stored at this offset from the start

label (i.e. for if we wanted to 'jump' to a label))

Some were easy, the very first line of a decompiled script (besides decompiler headers) looks something like:

The only parameter this command has is a reference to a label, so this is most likely (and in fact is) a jump statement, so we know all 0002s are jumps. Of course, finding what OpCodes do (and in fact finding the original number of parameters took a while to confirm) takes time, you have to have an idea first and then have to test your theory - many OpCodes have still not been named, but with the amount of OpCodes discovered so far, we have a general idea on what the mission script does.

Once the mission script had been cracked, people could write programs to read through it and output it in a form we could understand (based on a format of opcodes, text to say what they do and a list of parameter values - nothing like the original - the opcodes are needed to determine which opcode it is, the describing text is completely ignored). Originally there were two main decompilers, BWME (Barton Waterduck's Mission Editor) and CyQ's disassembler, each with their own compilers (to compile the decompiled code back into an SCM file). BWME quickly became the most commonly used, especially among newer coders, probably due to the fact that the parameters were inter-mixed with the code, so you had something like:

00B1: is_car $car in_cube $lowerx $lowery $lowerz $upperx $uppery $upperz 0

As opposed to the gtaMa/DisAsm format:

is_car_in_cube $car, $lowerx, $lowery, $lowerz, $upperx, $uppery, $upperz, 0

(also note the lack of OpCode in the second example, this builder uses a lookup to find the opcode (if the function is known) instead of just quoting it)

Although you can't see much difference with that example, it can make a lot of difference. Since Barton left the modding community, Seemann created an even more versatile decompiler, the Sanny Builder. It has become the most popular mission builder.

The Tools

There are three main builders for GTA 3, VC, and SA, and one for LCS and VCS.

Mission Builder

Main article: Mission Builder (BWME - note: although BWME was a slightly different tool, I shall be referring to this as that):

This tool uses only OpCodes to compile the code, all the text on the line is ignored. Traditionally, it decompiles to a file called main.scm.txt, which is just a big text file with all the code in it, expanded to be readable. This tool is used by the vast majority of mission coders as it is the most abundant and as most source code online is written for it, most people use it, the more people use it, the more code there is for it, so the more people use it.

Code format

Early builders used data type identifiers on all numbers, these were

 ? - small int (data type 04)
& - medium int (data type 05)
&& - big int (data type 01)<- global jump (data type 01)<- mission jump (data type 01 - negative addressing)
! - integer (data type 06)
$ - global variable (data type 02)
@ - local variable (data type 03)
?? - DMA global variable (like a global variable, only its memory position is its name, not assigned to it - data type 02)
: - label (text directly after used to reference this label in jumps)
"" - string (no data type, first 8 bytes after opcode when compiled)
# - model identifier (means you can enter the id name of a model rather than the number - data type 05 for the compiled number)

Later versions of the builder got rid of number type definitions, assigning types based on the size of the number. Integers were made integers by being not a whole number (e.g. 10.5 or 10.0 if you want a whole number defined as an integer). They also replaced DMA variables with global variables where the name was the hex address in decimal divided by 4 (each variable uses 4 bytes of memory).

Advantages

Widely used.
Commands related to the parameters.
Macros and program execution facilities inbuilt.

Disadvantages

Creator retired (no future updates / bug fixes).
Decompilation bugs (especially in certain advanced jumps).
Many unofficially usable SCM features uncatered for (although these are mostly advanced problems never experienced by the average coder).
Inconvenient syntax.

Other notes

GUI.
Compiler inbuilt.

Point

Main article: Point This is still very much in the development stages, it is the first user made high-level scripting tool (in fact, first high-level at all, even Rockstar's compiler is only an advanced parser) made for coding GTA. Originally developed for VC/III, this tool has been expanded to work for all three 3D games. One major disadvantage to this is that the file headers it writes, while readable by the game, are unrecognized by any line by line decompilers. So once a file has been compiled in Point, it cannot be decompiled again by another tool to see the exact generated code.

Sanny Builder

Main article: Sanny Builder.

Another new tool by Seemann. This is being developed for SA initially (although porting to other games may be an option) and is loosely based on BWME code. It was originally created using the SAscm.ini file from BWME 0.33 SA but has since been expanded to include many different features, some taken from Gtama, some new (such as a basic class system and direct HEX input (as requested by Y_Less)). This tool is still being developed and has now introduced a basic class system so you can do things such as "player.health += 5", this system is also being adapted for point.

The code format is based on a combination of both Gtama and BWME formats, although you can't force data types as you could in early Mission Builders (e.g. 0004: $var = 0&& which would normally be assigned one byte, not four).

Code format (note these are for SA, not VC as the others listed are)

$ - global variable
s$ - global string variable
v$ - global long string variable
@ - label (text directly AFTER used to reference this label in jumps)
@ - local variable (number BEFORE denotes variable)
@s - local string variable (number BEFORE denotes variable)
@v - local long string variable (number BEFORE denotes variable)
'...' - string (first 8 bytes after opcode when compiled)
"..." - debug string text
# - model identifier (means you can enter the id name of a model rather than the number)

Disassembler/Assembler

Main article: gtama (GTA Mission Assembler - gtaMa, Vice City Disassembler - DisAsm):

These tools use one word commands, although they may consist of multiple words concatenated by an underscore ("_"), e.g. is_player_defined. They still compile each line as-is (i.e. no interpretation or code generation) so the game will execute exactly the commands you enter. This is similar to programming in ASM mnemonics, whereas BWME is more similar to machine code. The decompiled file is split up into a number of .gsr files, each one containing the code to one mission. This reduces file sizes considerably as BWME generated files are huge (around 6 MB .txt files), containing the whole code. The code is in the format of a command, followed by a list of parameters, separated by spaces - this can make named variables easy to distinguish from commands.

The disassembler (DisAsm) is written by CyQ and the assembler (gtaMa) is written by Dan Strandberg. These two tools work together to de- and re-compile the code.

Code format

$ - global variable (data type 02)
! - local variable (data type 03)
@ - label (text directly after used to reference this label in jumps)
"" - string (no data type, first 8 bytes after opcode when compiled)
% - model identifier (means you can enter the id name of a model rather than the number - data type 05 for the compiled number)

Advantages

Small files sizes.
Clearer code - data and commands separated.
Active creator (although no longer developing).
More support for advanced features (supports memory hacking methods not widely used).
Open source.
Online updateable ini.
Format used on custom error handler for VC.

Disadvantages

Not widely used.
Code spread across multiple files - harder for searching.
Data not easily related to code.

Other notes

Command line based.