[Soot-list] Exception and Method Analysis

[Patrick Lam]

On 02/11/11 11:25, Richard Warburton wrote:
>>> So in a nutshell I think this will require a bit of manual labor but
>>> it should certainly be doable. Whether it's efficient in the end
>>> depends on the perspective I would say. Whole-program analyses like
>>> the one you are proposing tend to take a few minutes for realistic
>>> programs. That's just a matter of intrinsic complexity of the problem.
>>
>> I think that a substantial redesign of Soot could reduce this analysis
>> time significantly, but I'm not sure how far. It would sure be fun if
>> someone did that. Certainly with Soot's current architecture you're
>> looking at runtimes in minutes.
>
> Were there any particular changes you had in mind?

I don't have data. However, I've always suspected that creating Jimple 
for the whole program takes a long time, and one could probably avoid 
this while doing pointer analysis by directly analyzing the bytecode for 
the libraries and extracting the information that you'd need. There's 
also the possibility of parallelizing the Jimplification process, which 
I had started at some point, but never got around to finishing.

pat