[Array access detection] Consider address-of operation as a valid base candidate

[mari-mari]

Proposal

Currently, we consider expressions *(base+offset) as an array element access base[offset/type_size] if the base variable has type Pointer and offset satisfies valid offset requirements.

For instance, here:

void func(long * arr, size_t size){
for (int i=0; i<size; ++i)
     *(arr + 8*i) = ...;
...
}

we recognize *(arr+8*i) as arr[i].

However, we do not recognize &STUDENTS[var_2] here: gradebook.zip

$ python decompile.py gradebook add_student

extern void STUDENTS = 0;
...
        *(var_2 * 8 + &STUDENTS) = var_4;
...

Here address-operation (or reference) (&STUDENTS) is not recognized as valid base (pointer to the first element), as it is not a variable of type Pointer.

Since reference to a variable and a pointer to that variable are semantically equivalent, we should consider address operation as a valid base candidate.

Approach

Update and rename _is_pointer_variable function in ArrayAccessDetection to not only return True, if an expression is a variable of type Pointer but also if it is address UnaryOperation.

Also update getting array type -> in case as above type of unary operation operand can be used.

[mari-mari]

The expression &STUDENTS in *(var_2 * 8 + &STUDENTS) is the result of the bug during liffing of Binary Ninja DataVariable. The correctly lifted expression should be *(var_2 * 8 + STUDENTS). #106 should deal first with the bug.

Blocked by #106

[mari-mari]

The change proposed by the issue is not correct, since it is based on incorrect lifting (see comment above). The lifting was fixed and this change is not required and does not make sense. Closing.