[Soot-list] Phantom Classes and CFGs

[Bodden, Eric]

Hello.

> Hello,
> 
> I have read a paper that was just published from Gomes et al.
> http://www.nada.kth.se/~dilian/Papers/fase14.pdf
> 
> Essentially, their article is about over-approximating the call targets 
> and exceptional edges in the CFG when a method is unknown. This supports 
> incremental compilation as a side-benefit.
> 
> I was wondering how Soot is handling the same lack of information 
> (Phantom Classes). Are we also injecting conservative exceptional edges? 
> Are we already doing all this???

I don't have access to the paper right now. Soot's CFG edges are conservative w.r.t. checked exceptions, as those are declared. (this is true for ExceptionalUnitGraph, not for BriefUnitGraph) For unchecked exceptions, if I am not mistaken, then ExceptionalUnitGraph also conservatively adds edges from before any call statement to the respective catch blocks. This is certainly conservative but not maximally precise.

Anyway, sounds quite incremental by your description...

> P.S. They also boast that they are faster than Soot without giving any 
> experimental results. I'm going to stop now, before I say any bad words.

Odd to see that accepted at FASE.

Eric
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 841 bytes
Desc: Message signed with OpenPGP using GPGMail
Url : http://mailman.cs.mcgill.ca/pipermail/soot-list/attachments/20140311/2b4225e6/attachment-0001.bin