[Soot-list] SootResolver.bringToX...

[Patrick Lam]

On 12/07/2011 01:05 AM, Eric Bodden wrote:
> Hi all.
>
> Today it's me who has a question :-)
>
> I am having some students working on the SootResolver, and by looking
> at the code I am getting more and more confused. I always thought that
> the design was quite simple:
>
> - bringToHierarchies loads a class up to its super classes and interfaces
> - bringToSignatures loads methods and field signatures
> - bringToBodies fills in the bodies
> - All three methods load dependency classes to the required (lower levels).
>
> But apparently, this is not quite the case. For instance, we find this
> code in bringToSignatures:
>
>      private void bringToSignatures(SootClass sc) {
>          if(sc.resolvingLevel()>= SootClass.SIGNATURES ) return;
>          bringToHierarchy(sc);
>          sc.setResolvingLevel(SootClass.SIGNATURES);
>
>          for( Iterator fIt = sc.getFields().iterator(); fIt.hasNext(); ) {
>              final SootField f = (SootField) fIt.next();
>              addToResolveWorklist( f.getType(), SootClass.HIERARCHY );
>          }
>     ...
>
> As you can see here, this method iterates over all fields, which means
> that at this point it assumes all field signatures to already be
> present. In fact, the method does actually not do very much at all,
> except adding dependencies to the worklist. In particular, it does not
> really bring a class to signature level (except for changing the
> resolving level).
>
> Does anyone remember what was the intuition behind this?

I don't really remember who rewrote the SootResolver, but it wasn't me.

I do remember some exposure to this code thouogh. I think that there's a 
surprising amount of work that gets done while the Jimple is being 
created (e.g. by Coffi).

Whenever Coffi loads a class, it will create the SootFields etc for that 
class. (See soot.coffi.util.resolveFromClassFile).

It looks to me like whenever Soot loads a class to the SIGNATURES level, 
then the classes for the types of fields of that class will exist. Look 
at the doc comment for the bringToSignatures method: to me, it says that 
the signatures already exist.

My guess, then, is as follows:

If class c is at level:
HIERARCHIES: there exists a SootClass for c
SIGNATURES: ... and all of the types for the fields of c also exist
BODIES: ... and the bodies also exist (I can no longer recall where they 
get created, but it's somewhere else.)

pat