Reorder Analysis Dependencies

[fabianbs96]

The current definition of IFDSTabulationProblem and IDETabulationProblem has two drawbacks that are addressed in this PR:

• IDETabulationProblem inherits from IFDSTabulationProblem which requires the IFDSSolver to create a IfdsToIDETabulationProblem wrapper for propagating it to the generic IDESolver. IN thie PR, we flip the inheritance relation making IDETabulationProblem a superclass of IFDSTabulationProblem avoiding the mentioned issue.
• Each IfdsIde analysis problem requires (through their superclass) the complete set of IRDB, ICFG, PT and TH, from which most analyses only need the IRDB. Especially the PT computation oftentimes leads to a significant performance overhead, so we want to skip it, if we don't need it. However, as of now, deciding which analysis problem really needs which helper analysis requires in-depth knowledge of the implementation details of the analyses in question. In this PR, we let the IDE/IFDS TabulationProblem only require the IRDB such that each client analysis can declare itself, which helper analyses it needs

[pdschubert]

Thanks, @fabianbs96. I'll review this through the next few days :+1:

[vulder]

One quick question before hand. Isn't IFDS <: IDE conceptually the wrong direction? As a IDE problem is an IFDS one but not the other way around? Maybe a template structure where the "base parts" are IFDS and IDE parts get's woven in with a bit template magic. Could need a small adaption to the IDESolver, but maybe lead to a cleaner interface?

PS: I'm not sure, I'm just asking questions to explore design possibilities.

I'll also take a peak at the code later ;)

[fabianbs96]

Thanks @vulder for this advice; I was also thinking about that. However, as the IDESolver is designed as a base-class of the IFDSSolver it only makes sense for the IDEProblem to be a base-class of the IFDSProblem. I aggree with you that conceptually it would make more sense for the IFDS Solver to be the base of the IDESolver. The reason I changed it this way now is that I didn't want to do larger changes on the IDESolver. Anyway I would really like to see a code peak from you.

[vulder]

:+1: For the code, can you give me a few pointers to the important files, so I don't have to look at all the dependent changes :laughing:

[fabianbs96]

The most important files are the IDE- and IFDSTabulationProblem.h and maybe the second ctor of the IDESolver. I also did quite a lot changes in the AnalysisController, but these changes should not be that important on a conceptual level. To see how the changes affect analysis problems you may look at IFDSTaintAnalysis.h/.cpp