GSoC: Michael Madsen

Game script (bytecode) decompiler

First disassembler done!

The SCUMM disassembler is now complete, completing the second milestone.

As mentioned in my previous post, the next step is to write documentation on this, and then it's on to the code flow graph.

Posted in Decompiler, GSoC, SCUMM at 2010-06-06 17:15:59


After a brief discussion with my mentors, I'm going to use a different approach than I'd originally envisioned for the code generation part of the decompiler.

Basically, I'm going to try and emulate the principle used by descumm for generating the SCUMM code. This approach should be simpler than what I had originally envisioned, but still sufficiently generic to work without too many changes for KYRA.

Since this approach is vastly different from the original plan, I'm also re-evaluating my milestone plan. The original plan puts the expected date for SCUMM code generation at July 24, but I think I'll be able to get this done a few days earlier than that - I'm guessing it'll be around July 21 instead, so I'm setting that as the new date for that milestone for now, although it is still a pretty rough estimate. In a day or two, I will hopefully have a much better feel about how long this will really take, and then I will look at changing the other milestones as well - and possibly re-evaluating this one as well.

Posted in Decompiler, GSoC, SCUMM at 2010-07-13 21:22:13

First output!

I finally have some actual output working. It's only a proof of concept right now, but it's a nice start.

Here is the output the decompiler generates for one of the short Sam & Max scripts - one of the scripts I used for the code flow testing (link points to the graph for that script):
C:\scummvm\gsoc2010-decompiler>decompile -escummv6 script-33.dmp
var173 = var177;
var177 = 874;
Unknown opcode 6B99 at address 00000018
Unknown opcode 66 at address 0000001A
It only does these simple assignments for now - everything else is left unknown - but like I said, it's a start.

The opcodes it complains about are cursorCmd_Image and stopObjectCodeB, since I don't handle those yet.

Posted in Decompiler, GSoC, SCUMM at 2010-07-19 23:11:29

SCUMM over!

I've finished the SCUMM code generation now. Unfortunately, a bunch of stuff got in the way, and I didn't quite make my new revised deadline - but at least I finished in time for the original deadline.

There may still be some tweaks here and there which would be nice to add, but the code generation is working for all opcodes, and generating some pretty nice output. To give an example, here is the complete output for the script I showed you last time:
C:\scummvm\gsoc2010-decompiler>decompile -escummv6 script-33.dmp
00000000: VAR_GAME_LOADED = 0;
00000006: var173 = var177;
0000000C: var177 = 874;
00000012: cursorCmd_Image(var177, 93);
0000001A: stopObjectCodeB();
And just to show it does more than just such simple scripts, here's part of another, more complicated script:
C:\scummvm\gsoc2010-decompiler>decompile -escummv6 script-30.dmp
00000000: localvar2 = 1;
00000006: while ((localvar2 <= (60 * var132))) {
00000014:   breakHere();
00000015:   delaySeconds(1);
00000019:   if (((VAR_VIRT_MOUSE_X != localvar0) || (VAR_VIRT_MOUSE_Y != localvar1))) {
0000002B:     localvar0 = VAR_VIRT_MOUSE_X;
00000031:     localvar1 = VAR_VIRT_MOUSE_Y;
00000037:     jump 0;
0000003A:   }
0000003A:   if (bitvar1) {
00000040:     bitvar1 = 0;
00000046:     jump 0;
00000049:   }
00000049:   ++(localvar2);
0000004F: }
Now that all parts of the SCUMM decompilation have been handled, it's time to move on to the next engine - KYRA. My next task is therefore to get a disassembler for that written.

Posted in Decompiler, GSoC, SCUMM at 2010-07-24 23:09:43