decompiling back to C++?

[Shub-Nigurrath]

Hi,
have you ever heard of Desquirr? An IDA plugin, free, sourceforge, which tries to do exactly what you are talking about..

hxxp://desquirr.sourceforge.net/desquirr/

Desquirr is a decompiler plugin for IDA Pro. It is currently capable of simple data flow analysis of binaries with Intel x86 machine code.

might give some nice ideas, beside Boomerang, another free decompiling tool, already mentioned in this forum..

Originally Posted by Shub-Nigurrath

...have you ever heard of Desquirr?...
...Boomerang, another free decompiling tool...

There are a lot that creates pseudo C code from asm, because there is less information that is lost.
Ok, usally a C code is also a C++ code, but this is not what we want.
We know that the sofware has been written in C++, so we would like to get back the extra information that is in the source code.

There are some key points (vptr, dynamic_cast, static initialization...) that makes it possible to discover some of the original C++ structures.

I haven't heard about a tool that tries this.

/*
Just an example for a code that works differently in an old version of C and in C++ :

int a = 2 //**/ 2
;

*/

Hi all,

I found this thread by accident, maybe its not that old..

My first question:
why are you afraid of optimizations? I think that a decompiler shouldn't look for compiler-specific structures or patterns in code. Instead, it should read the _semantics_ of the code. The meaning of it. Everything that the program does is written down in the code. It will do the same whether it is optimized or not (or if not, then the difference is not important from the result's point of view). Imagine a simple intermediate language, that is close to asm, but with simpler constructs, mov, cmp, jumps, simple arithmetical and logical funtions. The transformation frop asm to this language is trivial. Then, since you know the meaning of c constructs, you can automatically find them in this language, you just have to map them there. Of course, there will be different constructs that mean the same, but why should we care.. like you do a loop with a for or with a while and some init code.
I think, that this approach could be used for this purpose.

The second... not a question:
Someone wrote about a c++ Vector or Boost template.. how you could reverse them. Well, as you may know, all the template stuff is like #define-s in a more advanced way. You can even instruct the compiler to give you the source after substituting everything but before compileing it. If you have the source of the templates that are possibly used in the code you wanna decompile, then you can parse the decompiled source, and look for the constructs that could be created with a template, and simply transform them back.

Ok, it's not that easy, I know. I just thinked a lot about this thing, but was too lazy to code anything... I just want to argue with you, maybe we'll come up with useful ideas in this discussion.

kp