[Soot-list] R: Spark generating rta call-graph:java.lang.OutOfMemoryError: GC overhead limit exceeded

Dario Amoroso d'Aragona darioamorosodaragona at gmail.com
Wed Sep 19 08:32:23 EDT 2018


Hi Eric, thank’s for your reply.
I have two projectst: p and p1. In p i have 5 classes with some methods with empty body and just someone with operation on primitive type. P1 is the same of p, but in one methods there is an arraylist. If i use rta for p i have a perfect complete call-grah but for p1 i have the problem that i have described before.
Yesterday, i tried to use spark with “on-fly-cg:true” and “rta:false” but, with this options, for p i have a call-graph with missing edges (maybe becasue i use non-static methods as entry-points), for p1 a complete perfect call-graph. 
Can the solution be Paddle? (But for now i not have succeeded to configure it ). 
Inviato da Posta per Windows 10

Da: Eric Bodden
Inviato: mercoledì 19 settembre 2018 14:11
A: Dario Amoroso d'Aragona
Cc: Soot list
Oggetto: Re: [Soot-list] Spark generating rta call-graph:java.lang.OutOfMemoryError: GC overhead limit exceeded

Hi Dario.


With the debugger i see that soot in generating call graph load all java method for ArrayList. 

That is the standard behaviour. Do you wish to exclude JDK classes from the analysis? If so, please try the option -no-bodies-for-excluded


The program run for 1 hour and than give me that error. I run it with java vm option -Xmx10g. 
I try to delete the two methods "realFailureTest" and "realPasstest" and all it's ok. I try to generate the spark call-graph with "rta" option false and "on-fly-cg" option true and all it's ok. Why?! 

I also noticed that you had  "on-fly-cg” set to false. This usually never makes sense. In this case, Soot first tries to construct an entire CHA-based call-graph for your entire code, and then runs its pointer-analysis on top of that callgraph. This is both less precise and less efficient than the on-the-fly mode. Hence  "on-fly-cg” should always be set to true.

Cheers
Eric 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.CS.McGill.CA/pipermail/soot-list/attachments/20180919/82ce9b06/attachment.html>


More information about the Soot-list mailing list