--removedeadswitchcases

Imagine a switch like this:

      switch (1) {
      default: {
        return false;
      }
      case 1:
        return true;
      }

or

    switch (boolean) { //illegal in java, legal in bytecode
    case 0:
      return false;
    case 1:
      return true;
    case 2:
    case 3:
    case 4:
    ...
      //never executed, remove it from the graph
      //an obfuscator could convert ifeq / ifne to switch and then add redundant extra cases
      //imagine a goto random label for each extra case
    default:
      //also not executed if both cases are handled
    }

I think there can be done a lot of optimization / obfuscation removal. Also not bad for a second pass.

[leibnitz27]

The former I can see as realistic, though I doubt anyone’s going to bother with an obfuscation so trivial. (Unless it’s @Marcono1234 generating psychotic switch expression examples)

The latter I’m really not keen on. We simply dont know at the bytecode level if something is a bool or not - you could declare it a bool parameter then assign 3 - while I try to infer the type/domain, it’s very possible to get it wrong.

(As an aside, abused booleans can lead to different results in different jvms, I have some tests somewhere I believe)

[leibnitz27]

Ok - I’m going to close this for now. There’s potentially some value in trying to add a last gasp recovery pass to do this, but it’s certainly not the sort of thing you’d want to be doing normally.