Fix bug with dataset iterator processing

[khatchad]

• Fix bug with iterating over datasets whose elements may be tuples.
  • This happens when a tensor dataset generator outputs tensors in a specific format.
  • Previously, Ariadne was not inferring tensors that were being picked out of the tuple during dataset iteration.
  • Added corresponding tests.
• Don't consider exceptions as tensor dataflow sources.
  • This was happening in the old code. I've added it to the new code as well.
• Since invocation instructions are processed in multiple places, I've extracted a common method.
  • The two different call sites differ on the src points-to variable to be added.
  • This guarantees that the order in which the points-to variables are processed doesn't matter.

[msridhar]

One suggestion

[msridhar]

Maybe processInvokeInstruction?

[khatchad]

Sorry for the delay. Yes, I also initially thought of that. But, then I decided to use method overloading:

  private static boolean processInstruction(
      SSAAbstractInvokeInstruction instruction,
      ...

In other words, the first parameter's type implicitly informs that an invoke instruction is being "processed." I can change it back if desired.

[msridhar]

My preference is to change the method name to processInvokeInstruction for a bit more clarity (in my opinion). But it's up to you; you read this code a lot more than I do 🙂 Not a huge deal

[khatchad]

I think that this is a good suggestion, but I wonder if we change this one, how many other ones are in a similar boat :).