Support for customizable/non-random search space traversal

[AlgorithmsAreCool]

Howdy,

Currently CsCheck only supports the sealed PCG class for it's source of generation. But I would like to replace the PCG type with an interface or abstract class allowing the search pattern to be customized. Basically, I want to not use random search all the time.

[AnthonyLloyd]

Hi,
I'd need to check there's no performance impact. We currently don't have complete performance metrics and it would be good to add.
How are you think of using a different RNG?

[AlgorithmsAreCool]

My hunch is that there would be a small impact at least because of a move to virtual calls, but in practice i would expect the time spent evaluating the SUT would dominate the cost of the RNG or other traversal strategy.

How are you think of using a different RNG?

I'm actually not wanting to use an RNG at all, but something more like a space filling curve or other monotonic function for traversing the search space. The design of the interface might be too limited to actual enumerate dimensions for a SFC to work, but basically, something non-random.

[johanlovgrenumc]

I am curious whether generalizing the source of generation could be used to replace the RNG with a fuzzer, as I am interested in combining fuzzing with property based testing. I had a master student try that out in .NET last year using FsCheck and sharpfuzz. It was a cool project but integrating all the different parts was quite complicated. Would be interesting if CsCheck could ease that integration!

[AnthonyLloyd]

@johanlovgrenumc sounds good. I'm not sure how it would work exactly. Happy to help with any questions or to look at code or issues you have.