[Soot-list] Potential bug fix for ThreadLocalObjectsAnalysis

[Khilan Gudka]

Sorry, i should also say, the outcome is the same if the x.f = new Object()
assignment is moved to after the start() call.

Khilan

On 15 March 2011 16:32, Khilan Gudka <khilan at doc.ic.ac.uk> wrote:

> Hi,
>
> Actually there is another bug. Here is my test case:
>
> public class SimpleSharing {
>
>     Object f;
>
>     public static void main(String[] args) {
>
>         final SimpleSharing x = new SimpleSharing();
>
>         x.f = new Object();
>
>         // share x
>         (new Thread() {
>             public void run() {
>                 x.f = null;
>             }
>         }).start();
>     }
>
> }
>
>
> TLO says that x is thread-local and x.f is shared but x is shared here as
> well?
>
> Thanks,
> Khilan
>
> On 15 March 2011 13:33, Khilan Gudka <khilan at doc.ic.ac.uk> wrote:
>
>> Hi,
>>
>> I think I have a potential bug fix for
>> http://www.sable.mcgill.ca/pipermail/soot-list/2009-September/002504.html.
>> It seems that the bug is due to an assumption that if an object is
>> thread-local then all it's fields are also thread-local. This manifests
>> itself in ClassLocalObjectsAnalysis.java with the call to
>> setAllFieldsLocal() in the block:
>>
>> else if( SmartMethodLocalObjectsAnalysis.isObjectLocal(dfa,
>> containingMethod, containingContext, iie.getBase()) )
>>
>> {
>>
>> // calling a method on a local object
>>
>> callingContext.setAllFieldsLocal();
>>
>> callingContext.setThisLocal();
>>
>> }
>>
>>
>> However, Arie's test-case breaks this assumption because although publish
>> is thread-local, publish.d isn't. Hence, I commented out the call to
>> setAllFieldsLocal() and it is now giving the correct result, that publish is
>> thread-local and publish.d is thread-shared.
>>
>>
>> Please let me know if this is correct.
>>
>>
>> Cheers,
>>
>> Khilan
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.mcgill.ca/pipermail/soot-list/attachments/20110315/3831e71f/attachment-0001.html