[Soot-list] bug in the Jimple parser

Guillaume Salagnac guillaume.salagnac at imag.fr
Tue Sep 27 03:44:21 EDT 2005


On Mon, 26 Sep 2005, Patrick LAM wrote:

> Guillaume Salagnac wrote:
> 
> > I was trying to improve Soot's whole-program running time, which is more
> > than two minutes here, so i tried to make it load pre-generated jimple
> > files instead of parsing .class files.
> 
> Please let us know if that helps.  Originally I was hoping that the Jimple
> parser would make things faster, but I didn't observe any speedup in
> practice, especially in single-file mode.

Neither did I.

I'm working in whole-program mode, and the approach I was taking was to 
generate .jimple files for the whole classpath, hoping that soot would 
load them faster than their .class counterparts.

So, after I found some classes to cause Soot to crash, I removed the 
corresponding .jimple files (Integer, Math, StrictMath and a few others) 
and repeated the whole thing. Almost all the classpath was loaded from 
.jimple, except for those classes for which soot was falling back to 
.class.

But it was not faster at all. Soot runs in about 2 minutes for a toy 
example (a HelloWorld) with the 1.4.2 jdk classpath.


java -mx800m soot.Main -include-all -w -app -f J HelloWorld
...
Soot has run for 1 min. 58 sec.


java -mx800m soot.Main -include-all -w -app -f J -src-prec J HelloWorld 
...
Soot has run for 1 min. 58 sec.



Too bad, 
-G

--
Guillaume Salagnac
PhD student, Vérimag Laboratory, Grenoble, France.
Je sers la science et c'est ma joie.


More information about the Soot-list mailing list