I've fixed many issues in this codebase, among other things - memory reallocation during decompilation.

To reflect those fixes, I've edited the original readme a bit.

dcc Distribution

The code provided in this distribution is (C) by their authors:

• Cristina Cifuentes (most of dcc code)
• Mike van Emmerik (signatures and prototype code)
• Jeff Ledermann (some disassembly code)

and is provided "as is". Additional contributor list is available on GitHub.

The following files are included in the dccoo.tar.gz distribution:

• dcc.zip (dcc.exe DOS program, 1995)
• dccsrc.zip (source code .c, .h for dcc, 1993-1994)
• dcc32.zip (dcc_oo.exe 32 bit console (Win95/Win-NT) program, 1997)
• dccsrcoo.zip (source code .cpp, .h for "oo" dcc, 1993-1997)
• dccbsig.zip (library signatures for Borland C compilers, 1994)
• dccmsig.zip (library signatures for Microsoft C compilers, 1994)
• dcctpsig.zip (library signatures for Turbo Pascal compilers, 1994)
• dcclibs.dat (prototype file for C headers, 1994)
• test.zip (sample test files: .c .exe *.b, 1993-1996)
• makedsig.zip (creates a .sig file from a .lib C file, 1994)
• makedstp.zip (creates a .sig file from a Pascal library file, 1994)
• readsig.zip (reads signatures in a .sig file, 1994)
• dispsrch.zip (displays the name of a function given a signature, 1994)
• parsehdr.zip (generates a prototype file (dcclibs.dat) from C *.h files, 1994)

Note that the dcc_oo.exe program (in dcc32.zip) is a 32 bit program, so it won't work under Windows 3.1. Also, it is a console mode program, meaning that it has to be run in the "Command Prompt" window (sometimes known as the "Dos Box"). It is not a GUI program.

The following files are included in the test.zip file: fibo, benchsho, benchlng, benchfn, benchmul, byteops, intops, longops, max, testlong, matrixmu, strlen, dhamp. The version of dcc included in this distribution (dccsrcoo.zip and dcc32.exe) is a bit better than the first release, but it is still broken in some cases, and we do not have the time to work in this project at present so we cannot provide any changes. Comments on individual files:

• fibo (fibonacci): the small model (fibos.exe) decompiles correctly, the large model (fibol.exe) expects an extra argument for scanf(). This argument is the segment and is not displayed.
• benchsho: the first scanf() takes loc0 as an argument. This is part of a long variable, but dcc does not have any clue at that stage that the stack offset pushed on the stack is to be used as a long variable rather than an integer variable.
• benchlng: as part of the main() code, LO(loc1) | HI(loc1) should be displayed instead of loc3 | loc9. These two integer variables are equivalent to the one long loc1 variable.
• benchfn: see benchsho.
• benchmul: see benchsho.
• byteops: decompiles correctly.
• intops: the du analysis for DIV and MOD is broken. dcc currently generates code for a long and an integer temporary register that were used as part of the analysis.
• longops: decompiles correctly.
• max: decompiles correctly.
• testlong: this example decompiles correctly given the algorithms implemented in dcc. However, it shows that when long variables are defined and used as integers (or long) without giving dcc any hint that this is happening, the variable will be treated as two integer variables. This is due to the fact that the assembly code is in terms of integer registers, and long registers are not available in 80286, so a long variable is equivalent to two integer registers. dcc only knows of this through idioms such as add two long variables.
• matrixmu: decompiles correctly. Shows that arrays are not supported in dcc.
• strlen: decompiles correctly. Shows that pointers are partially supported by dcc.
• dhamp: this program has far more data types than what dcc recognizes at present.

Our thanks to Gary Shaffstall for some debugging work. Current bugs are:

• [ ] if the code generated in the one line is too long, the (static) buffer used for that line is clobbered. Solution: make the buffer larger (currently 200 chars).
• [ ] the large memory model problem & scanf()
• [ ] dcc's error message shows a p option available which doesn't exist, and doesn't show an i option which exists.
• [x] there is a nasty problem whereby some arrays can get reallocated to a new address, and some pointers can become invalid. This mainly tends to happen to larger executable files. A major rewrite will probably be required to fix this.

For more information refer to the thesis "Reverse Compilation Techniques" by Cristina Cifuentes, Queensland University of Technology, 1994, and the dcc home page: http://www.it.uq.edu.au/groups/csm/dcc_readme.html

Please note that the executable version of dcc provided in this distribution does not necessarily match the source code provided, some changes were done without us keeping track of every change.

Using dcc

Here is a very brief summary of switches for dcc:

• a1, a2: assembler output, before and after re-ordering of input code
• c: Attempt to follow control through indirect call instructions
• i: Enter interactive disassembler
• m: Memory map
• s: Statistics summary
• v, V: verbose (and Very verbose)
• o filename: Use filename as assembler output file

If dcc encounters illegal instructions, it will attempt to enter the so called interactive disassembler. The idea of this was to allow commands to fix the problem so that dcc could continue, but no such changes are implemented as yet. (Note: the Unix versions do not have the interactive disassembler). If you get into this, you can get out of it by pressing ^X (control-X). Once dcc has entered the interactive disassembler, however, there is little chance that it will recover and produce useful output.

If dcc loads the signature file dccxxx.sig, this means that it has not recognised the compiler library used. You can place the signatures in a different direcory to where you are working if you set the DCC environment variable to point to their path. Note that if dcc can't find its signature files, it will be severely handicapped.