[Soot-list] How can I tell if the two arrays overlap?

[Marc-Andre Laverdiere-Papineau]

Thinking about this, I see the following:

First: do you need to check in-memory overlap, or contents overlap?

My answers below assume the first. For the other, the collection overlap 
paper would be your friend.
1- Java Arrays cannot be referenced like in C. You cannot do something 
like &array[12]. So overlap is a somewhat useless notion.

2- So you need to check for equality. That can be done at runtime using 
the == operator, which checks against the handles AFAIR. That may not 
work well if the array is passed between methods.

Statically, I think, constant propagation or SSA form would answer the 
question. If the objects are the same based on a definitive analysis 
(intersection confluence operator, typically), then you'd have a good 
answer.

On 2012-07-29 23:19, John Dean wrote:
> Hi,
>
> If there are two array references with different names for the two arrays,
> how can I tell if the two arrays overlap (for example, one array is
> 2-dimensional, and the other array is a row inside the first array)?
>
> I've searched through the soot api for clues, but I haven't find anything
> suitable yet. If I was working with Java source code instead of Jimple, I
> would call arr1.containsAll(arr1), but there's nothing comparable in soot,
> as far as I can tell. I realize that soot is a static analysis tool, so it
> won't allow me to find an array element's value.
>
> I think I know how to retrieve the array references and end up with
> JArrayRef variables, and I can call JArrayRef's getBase and getIndex methods
> to get the array variable name and the index, respectively, but I don't know
> how that helps me.
>
> Maybe if I could somehow get the size of each array and get the starting
> address of each array, then I suppose I could use that information to tell
> if the arrays overlapped. That sounds wrong, but maybe not?
>
> Or maybe I should use each array's retrieved base and then search for
> definitions of the bases and "chase" the definitions back to see if they
> intersect. That sounds like a lot of work, and I don't know if it's even
> possible.
>
> Any advice would be greatly appreciated.
>
> Thanks,
> John
>
> _______________________________________________
> Soot-list mailing list
> Soot-list at sable.mcgill.ca
> http://mailman.cs.mcgill.ca/mailman/listinfo/soot-list
>


-- 
Marc-André Laverdière-Papineau
Étudiant au doctorat - PhD Student