Maintain `goto`s when present in the original C code

[frabert]

Currently, Rellic always avoids generating gotos, favoring ifs and whiles. This is not always favorable, e.g.

extern int doA(void);
extern void undoA();

extern int doB(void);
extern void undoB();

extern int doC(void);
extern void undoC();

extern void foo(void);

int main(void) {
  if(!doA()) {
    goto exit;
  }
  if(!doB()) {
    goto cleanupA;
  }
  if(!doC()) {
    goto cleanupB;
  }

  foo();

cleanupA:
  undoA();
cleanupB:
  undoB();
cleanupC:
  undoC();
exit:
  return 0;
}

roundtrips into

unsigned int main();
unsigned int doA();
unsigned int doB();
unsigned int doC();
void foo();
void undoA();
void undoB();
void undoC();
unsigned int main() {
    unsigned int val0;
    unsigned int val1;
    unsigned int val2;
    unsigned int var3;
    var3 = 0U;
    val0 = doA();
    if (val0 != 0U) {
        val1 = doB();
        if (val1 != 0U) {
            val2 = doC();
        }
        if (val1 != 0U && val2 != 0U) {
            foo();
        }
        if (!(val1 == 0U || val2 == 0U || val0 == 0U) || !(val1 != 0U || val0 == 0U)) {
            undoA();
        }
        undoB();
        undoC();
    }
    return 0U;
}

If debug metadata regarding labels is available, Rellic should probably stick to generating jumps instead.

[surovic]

I have a feeling this might be a significant issue, since the core concept of rellic's underlying structurization doesn't allow for unstructured control flow i.e. goto.

[pgoodman]

Perhaps there's a way to support this, i.e. via sequencing regions or something, where a goto itself is treated as some kind of black box. This is not a priority feature but one to keep in mind, as C code that uses gotos has often made a deliberate choice to use gotos, and the label names are semantically valuable. Being able to maintain that value seems advantageous.

[surovic]

...and the label names are semantically valuable. Being able to maintain that value seems advantageous.

Agreed.

While it might be possible to somehow incorporate goto statements into the code itself, I think it would be much better to somehow annotate blocks of code near the label with the name of the label or something like that. And then if the code doesn't look as nice, I'd try to figure out how to improve reaching condition simplification. This way we stay in line of the original algorithm and we improve non-goto outputs as well.