Re: Introduction

[Ondrej Lhotak <olhotak@sable.mcgill.ca>]

On Wed, Jun 02, 2004 at 04:58:06PM -0400, Patrick LAM wrote:
> On Wed, 2 Jun 2004, Helge Jensen wrote:
> 
> > The idea is to implement a relatively simple BTA for jimple by:
> >
> >    1. using structural analysis to recreate the compositional
> > control-flow from jimple-code into a simple "WHILE"-type language
> 
> Dava does this.  I think it's in the main Soot tree.
> 
> I think it shouldn't take 20 minutes to do a whole-program analysis, but
> I'm not current on that area.  Maybe Ondrej will comment on that.

I would guess most of those 20 minutes are being taken by the Java
garbage collector. On my Athlon 2500+ (1833 MHz), doing a whole-program
analysis of Hello World with j2sdk1.4.2 takes 57 seconds. This is if I
run java with -Xmx300m so that it has some breathing room on the heap
and isn't garbage collecting all the time.

To answer the original question, no, Soot doesn't have an easy way to
make it analyze only part of the program when it's doing a whole program
analysis.

As for the ConcurrentModificationException, I haven't been able to
reproduce it with the source file and the command line given. The
handleClass method should not cause new classes to be added to the
Scene. In fact, nothing in Spark should cause new classes to be added to
the Scene, because Spark relies on the Hierarchy, which will be wrong if
classes are added after it is built.

Ondrej

> 
> pat
> 
>