[Soot-list] (GeomPointsTo) Spark PTA Re-analysis

Fri Apr 5 01:10:53 EDT 2013

Hi, Michael:

I'm glad to hear that you solve the problem.

As I know, it's a long time the soot team plans to release soot 2.6,
because many bugs have been fixed since 2.5. Eric is careful of every
commitment to soot. Therefore, don't worry too much about the development
version quality.

And, the geomPTA is also updated significantly since 2.5. It's more
efficient and a bug for analyzing programs with more than 2^63 contexts is
fixed.

Cheers,
Xiao

On Wed, Apr 3, 2013 at 10:38 PM, Michael Gordon <mgordon at mit.edu> wrote:

> Hello Xiao,
>
> Thank you so much for your prompt help!  This indeed did fix the problem
> for me.  I will inspect the analysis result to make sure it is correct.
>  Funny, that I was experimenting with this global, but I must have released
> too much state in combination with it.
>
> Regarding your suggestion of using the development version, currently, I'm
> using Soot 2.5, the newest version on the soot downloads page.  Should I
> instead switch to the github version?  Is it stable?
>
> Best,
> Michael
>
>
> On Wed, Apr 3, 2013 at 4:41 AM, Richard Xiao <richardxx at cse.ust.hk> wrote:
>
>> Hi, Michael:
>>
>> Now I found the root cause.
>> The problem is because SPARK caches the analyzed methods in a global
>> field "G.v().MethodPAG_methodToPag". Therefore, if you don't clean the
>> cache, the methods are analyzed again in your second PTA. Since geomPTA
>> obtains program information from SPARK and SPARK contains no program
>> information in the second run (because SPARK skips all analyzed methods in
>> the cache), geomPTA crashes due to uninitialized variables. Just add a line:
>>
>> G.v().MethodPAG_methodToPag = new HashMap<SootMethod, MethodPAG>();
>>
>> before your "SparkTransformer.v().transform("",opt);", your problem is
>> solved. Please try it out.
>>
>>
>> Cheers,
>> Xiao
>>
>>
>>
>> On Tue, Apr 2, 2013 at 11:49 PM, Michael Gordon <mgordon at mit.edu> wrote:
>>
>>> Hello Xiao,
>>>
>>> I initially tried just to release the CG and the PTA, but it did not
>>> work.  Here is the error I am receiving:
>>>
>>> Exception in thread "main" java.lang.ArrayIndexOutOfBoundsException: -1
>>> at java.util.ArrayList.get(ArrayList.java:324)
>>> at
>>> soot.jimple.spark.geom.geomPA.OfflineProcessor.add_graph_edge(OfflineProcessor.java:541)
>>> at
>>> soot.jimple.spark.geom.geomPA.OfflineProcessor.buildInstanceAssignmentGraph(OfflineProcessor.java:147)
>>> at
>>> soot.jimple.spark.geom.geomPA.OfflineProcessor.runOptimizations(OfflineProcessor.java:85)
>>> at
>>> soot.jimple.spark.geom.geomPA.GeomPointsTo.solve(GeomPointsTo.java:995)
>>> at
>>> soot.jimple.spark.SparkTransformer.internalTransform(SparkTransformer.java:188)
>>> at soot.SceneTransformer.transform(SceneTransformer.java:39)
>>>
>>>
>>> At this point, my main method is quite large, and has a lot
>>> transformation.  I do not have a simple version to demonstrate this bug.
>>>  However, my consecutive calls to PTA look like this:
>>>
>>> ...
>>> runPTA();
>>> Scene.v().releaseCallGraph();
>>> Scene.v().releasePointsToAnalysis();
>>> runPTA();
>>> ...
>>>
>>> public static void runPTA() {
>>>
>>> HashMap opt = new HashMap();
>>>
>>> opt.put("geom-pta","true");
>>>
>>> opt.put("geom-encoding", "geom");
>>>
>>> opt.put("geom-worklist", "PQ");
>>>
>>> opt.put("geom-eval", "0");
>>>
>>> opt.put("geom-trans", "false");
>>>
>>> opt.put("geom-frac-base", "40");
>>>
>>> opt.put("geom-blocking", "true");
>>>
>>> opt.put("geom-runs", "1");
>>>
>>> opt.put("enabled","true");
>>>
>>> opt.put("verbose","false");
>>>
>>> opt.put("ignore-types","true");
>>>
>>> opt.put("force-gc","false");
>>>
>>> opt.put("pre-jimplify","false");
>>>
>>> opt.put("vta","false");
>>>
>>> opt.put("rta","false");
>>>
>>> opt.put("field-based","false");
>>>
>>> opt.put("types-for-sites","false");
>>>
>>> opt.put("merge-stringbuffer","false");
>>>
>>> opt.put("string-constants","true");
>>>
>>> opt.put("simulate-natives","true");
>>>
>>> opt.put("simple-edges-bidirectional","false");
>>>
>>> opt.put("on-fly-cg","true");
>>>
>>> opt.put("simplify-offline","false");
>>>
>>> opt.put("simplify-sccs","false");
>>>
>>> opt.put("ignore-types-for-sccs","false");
>>>
>>> opt.put("propagator","worklist");
>>>
>>> opt.put("set-impl","double");
>>>
>>> opt.put("double-set-old","hybrid");
>>>
>>> opt.put("double-set-new","hybrid");
>>>
>>> opt.put("dump-html","false");
>>>
>>> opt.put("dump-pag","false");
>>>
>>> opt.put("dump-solution","false");
>>>
>>> opt.put("topo-sort","false");
>>>
>>> opt.put("dump-types","true");
>>>
>>> opt.put("class-method-var","true");
>>>
>>> opt.put("dump-answer","false");
>>>
>>> opt.put("add-tags","false");
>>>
>>> opt.put("set-mass","false");
>>>
>>> SparkTransformer.v().transform("",opt);
>>>
>>> }
>>>
>>> Any thoughts on the above error?  I can work on creating a smaller main
>>> driver that recreates the problem if no one has suggestions.
>>>
>>> Cheers!
>>> Michael
>>>
>>> ---
>>>
>>>
>>> Hi, Michael:
>>>
>>> The problem is you release important information that is needed by
>>> points-to analysis. For example, "releaseFastHierarchy" destroys the type
>>> inheritance graph, which is used by either SPARK or Geom.
>>>
>>> Please just try releasing only the CallGraph and the PointsToAnalysis
>>> after
>>> your transformation. If it doesn't work neither, please show us your full
>>> main function that we can reproduce your failure.
>>>
>>> Cheers,
>>> Xiao
>>>
>>> --
>>> Michael Gordon, PhD
>>> Research Scientist, MIT CSAIL
>>> http://people.csail.mit.edu/mgordon
>>> Program Head, MIT AITI
>>> http://aiti.mit.edu
>>>
>>> _______________________________________________
>>> Soot-list mailing list
>>> Soot-list at sable.mcgill.ca
>>> http://mailman.cs.mcgill.ca/mailman/listinfo/soot-list
>>>
>>>
>>
>>
>> --
>> Richard Xiao Xiao
>> PhD Student @ CSE @ Hong Kong University of Science and Technology
>> www.cse.ust.hk/~richardxx
>>
>
>
>
> --
> Michael Gordon, PhD
> Research Scientist, MIT CSAIL
> http://people.csail.mit.edu/mgordon
> Program Head, MIT AITI
> http://aiti.mit.edu
>

-- 
Richard Xiao Xiao
PhD Student @ CSE @ Hong Kong University of Science and Technology
www.cse.ust.hk/~richardxx
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.mcgill.ca/pipermail/soot-list/attachments/20130405/b8ba24b8/attachment.html 