[Soot-list] my StrongLocalMustAliasAnalysis gives unexpected results

[Zhoulai]

We were talking about an example where thredefinition of variable 'x' is
not mandatory (see below the program) , but your StrongMustAliasAnalysis
associates with 'x' an integer , rather than UNKNOWN.

     > > ****************
     > > 1:     static void test5(int i){
     > > 2:         while *(i>0)*{
     > > 3:             A x = new A();             *// The re-assignment is
NOT mandatory here.*
     > > 4:             A y = x;                       //*but 'x' is bound
to an integer value. This is not sound!*
     > > 5:             String s = "hello";
     > >          }
     > >      }
     > > **************

'x' is *potentially* re-assigned in the loop
==> 'x' is expected to be bound to "UNKNOWN" by your
StrongMustAliasAnalysis   (**)

But, at line (5) your analyzer infers that x is bound to some integer
value  (*)

So the expected (**) and the implemented (*)  contradict.

Zell.


On Sun, Oct 21, 2012 at 1:04 PM, Patrick Lam <plam at sable.mcgill.ca> wrote:

> As I said, variable redefinition only gives UNKNOWN if the redefinition is
> not mandatory.
>
> pat
>
>
> On 10/21/2012 01:26 AM, Zhoulai wrote:
>
>> Thanks Patrick. I see your point. Don't forget that your
>> StrongMustAliasAnalysis, as described by your 2008's paper, is supposed
>> to consider variable redefinition in loops!
>>
>> For the k-th iteration of the loop, a new object o_k is created and
>> bound to x (which overwrites its earlier state) , so the "*states
>> history* "  related to x will be
>>
>> x->o_0, x -> o_2, x->o_k....
>> which mean that ***Over time*** x is bound to {o_0,o_1,o_2 ...}. I
>>
>> believe this is the way your StrongMustAliasAnalysis in your paper
>> reasons, right?
>>
>> My apologies for any misunderstandings.
>> Zell.
>>
>> On Sun, Oct 21, 2012 at 6:33 AM, Patrick Lam <plam at sable.mcgill.ca
>> <mailto:plam at sable.mcgill.ca>> wrote:
>>
>>     On 10/21/2012 12:25 AM, Zhoulai wrote:
>>      > > ****************
>>      > > 1:     static void test5(int i){
>>      > > 2:         while (i>0){
>>      > > 3:             A x = new A();
>>      > > 4:             A y = x;
>>      > > 5:             String s = "hello";
>>      > >          }
>>      > >      }
>>      > > **************
>>
>>      > By your specification in your paper, the strong representative only
>>      > must-alias itself at the same program location if it is not
>>     guaranteed
>>      > that this representative represents a single run-time object.
>>      >
>>      > Here variable 'x'' is associated with  more than 1 run-time
>>     objects.  It
>>      > is therefore expected to be bound to UNKNOWN.
>>
>>     x only points to one object at a time in this case. The only way that
>> x
>>     might point to more than one object is if there are two different
>>     definitions that reach x (perhaps from different iterations of the
>>     loop), like if you had this program instead:
>>
>>           A x = null;
>>           if (...) x = new A();
>>           A y = x;
>>
>>     Then you should not get UNKNOWN. But in the program you wrote, there's
>>     only one object that x could possibly contain.
>>
>>     pat
>>
>>     ______________________________**_________________
>>     Soot-list mailing list
>>     Soot-list at sable.mcgill.ca <mailto:Soot-list at sable.**mcgill.ca<Soot-list at sable.mcgill.ca>
>> >
>>     http://mailman.cs.mcgill.ca/**mailman/listinfo/soot-list<http://mailman.cs.mcgill.ca/mailman/listinfo/soot-list>
>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.mcgill.ca/pipermail/soot-list/attachments/20121021/0b3a4165/attachment-0001.html