[Soot-list] Using a string to find an edge

[Jason Ott]

Yes, there are better ways of getting a more accurate call graph.  But for
the purposes that we needed, PScout was the best choice -- it accomplished
much of the other necessary work we needed in this case.

On Mon, Apr 13, 2015 at 11:49 PM, Bodden, Eric <
eric.bodden at sit.fraunhofer.de> wrote:

> Hi Jason.
>
> Would you mind sharing this conversation with the Soot list? It  might out
> other people in the future.
>
> > Yes, I want to refine the call graph -- but not in the way I think
> you're thinking.  We have a call graph for the entire Android OS, created
> by analyzing the classes and weaving together the calls between classes.
> This is the way PScout (http://pscout.csl.toronto.edu/) builds the call
> graph.  Obviously, there are going to be inconsistencies and errors by
> doing it that way.  That is our starting point.  This is where the fan out
> is introduced.  I can properly detect it, which gives me a list of methods
> and their complete definition (except the implementation).  So then I run
> SPARK using PTA to construct a new call graph starting somewhere where the
> fan out begins.  Because I have a list of all the method definitions I
> have, and that I cannot create new a new unit, I was just going to compare
> my definitions with the definitions in the SPARK generated call graph.
>
> Ok, I see. Sounds like quite a complex approach.
>
> >
> > Once I have the corrected path, I can alter the PScout call graph to
> reflect the proper path.
> >
> > I understand the inefficiencies in doing it this way, but it seems like
> it's my only option.
>
> Would it not be possible to somehow use only the Soot-computed call graph?
> PScout is kind of outdated anyway and there are more accurate approaches
> that build on top of Soot.
>
> Cheers,
> Eric
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://mailman.CS.McGill.CA/pipermail/soot-list/attachments/20150414/8f891936/attachment.html