[Soot-list] PatchingChain.insertBefore(Chain<E> toInsert, E point): bug?

[Khilan Gudka]

Hi Eric,

I've just come across another issue, which the previous fix doesn't solve.
If the first unit u of the chain being inserted itself points to "point",
then the code still results in u then pointing to u. For example, inserting

label0:
if (x == 0) goto label1;

before
label1:
nop

would lead to:

label0:
if (x == 0) goto label0;

label1:
nop

I had a look and found that AbstractUnit.redirectJumpsToThisTo(Unit
newLocation) does not check if the jump source is itself "newLocation." Is
this the intended behaviour?

What do you suggest?

Thanks,
Khilan

2009/12/10 Eric Bodden <bodden at st.informatik.tu-darmstadt.de>

> Seems to cause no problems... This is now committed in revision 3351.
>
> Thanks again,
> Eric
>
> --
> Eric Bodden
> Software Technology Group, Technische Universität Darmstadt, Germany
> Tel: +49 6151 16-5478    Fax: +49 6151 16-5410
> Mailing Address: S2|02 A209, Hochschulstraße 10, 64289 Darmstadt
>
>
>
> 2009/12/10 Eric Bodden <bodden at st.informatik.tu-darmstadt.de>:
> > Hi Khilan.
> >
> > Thanks for the patch. This looks like a correct patch to an actual bug
> > to me. I will try it out with our test suite today.
> > Eric
> >
> > --
> > Eric Bodden
> > Software Technology Group, Technische Universität Darmstadt, Germany
> > Tel: +49 6151 16-5478    Fax: +49 6151 16-5410
> > Mailing Address: S2|02 A209, Hochschulstraße 10, 64289 Darmstadt
> >
> >
> >
> > 2009/12/9 Khilan Gudka <khilan at doc.ic.ac.uk>:
> >> Hi,
> >> The method PatchingChain.insertBefore(Chain<E> toInsert, E point)
> inserts
> >> the chain "toInsert" just before "point" in the receiver chain. It does
> this
> >> by calling PatchingChain.insertBefore(E toInsert, E point) for each Unit
> in
> >> a reversed version of toInsert. To keep the correct control flow,
> >> PatchingChain.insertBefore(E toInsert, E point) redirects jumps to
> "point"
> >> to jump to "toInsert" instead. However, I've found that this leads to
> bugs
> >> when inserting a chain that contains jumps itself. For example,
> inserting
> >> the jimple code:
> >>         if $r == 0 goto label0;
> >>         staticinvoke <mypackage.MyClass: void m()>();
> >>      label0:
> >>         staticinvoke <mypackage.MyClass: void n()>();
> >> Which corresponds to:
> >> if($r) {
> >>     MyClass.m();
> >> }
> >> MyClass.n();
> >> Would be turned into the following when using the
> >> above PatchingChain.insertBefore(Chain<E> toInsert, E point) method:
> >>      label0:
> >>         if $r == 0 goto label0;
> >>         staticinvoke <mypackage.MyClass: void m()>();
> >>         staticinvoke <mypackage.MyClass: void n()>();
> >> The reason being that when insertBefore(E toInsert, E point) is
> repeatedly
> >> called, the jump to "staticinvoke <mypackage.MyClass: void n()>();" is
> >> redirected to the first statement "if $r == 0 goto label0;".
> >> I think the solution is to, lastly, redirect jumps to "point" to the
> first
> >> unit in toInsert (here is a snippet from the bottom
> >> of PatchingChain.insertBefore(Chain<E> toInsert, E point) and the
> proposed
> >> changes are in bold):
> >>         E previousPoint = point;
> >>         Iterator<E> it = backwardList.iterator();
> >>         while (it.hasNext())
> >>         {
> >>             E o = it.next();
> >>             insertBeforeNoRedirect(o, previousPoint);
> >>             previousPoint = o;
> >>         }
> >>         ((Unit) point).redirectJumpsToThisTo((Unit)
> toInsert.getFirst());
> >> Does this make sense or have I overlooked something?
> >> Thanks,
> >> Khilan
> >> _______________________________________________
> >> Soot-list mailing list
> >> Soot-list at sable.mcgill.ca
> >> http://mailman.cs.mcgill.ca/mailman/listinfo/soot-list
> >>
> >>
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.mcgill.ca/pipermail/soot-list/attachments/20091212/9be543c7/attachment.html