reiddraper/simple-check

`Generator` protocol is awkward

Closed this issue · 2 comments

cemerick commented

FWIW, this is more an experience report + continuation of prior conversation than actual bug.

I've found the Generator protocol to be a bit of a PITA to use. I understand the utility and purpose of having each generator paired with its concomitant shrinker, but keep running into some semantic and practical issues:

  1. I find myself often wanting to use e.g. the same shrinker for multiple "types" of generators. e.g. I have a shrink-number fn (similar to halfs but it works with decimals as well as integers), which I want to use to shrink all numeric values, regardless of their concrete type or method of generation (i.e. random vs. weighted, etc). My options of pairing it up with them aren't really very good though: delegate from a reified Generator, or come up with a dummy type to which I can extend Generator with a map of functions. This seems like busywork, esp. if I'm aiming to reuse existing value-generating facilities.
  2. It seems that the "generator" concept is entirely captured by arbitrary, and shrink is tacked on simply because shrinkers need to be paired with generators. Perhaps this is nomenclature borrowed from Haskell's quickcheck?
  3. Looking ahead, it seems like arbitrary implies "fully random", as opposed to other generation strategies that might be geometric or weighted in nature. Does that mean that Generator will end up growing more methods?

To be clear, none of this prevents me from using simple-check, but it doesn't help me look forward to writing a new Generator. :-)

Without being familiar with quickcheck's nomenclature and background, it seems like each "parameter specification" provided to quick-check could just be a map with two slots, :generator and :shrinker, without complicating the implementation details of the lib particularly much.

reiddraper commented

Protocol changed as of 46359bd.

reiddraper commented

Looking ahead, it seems like arbitrary implies "fully random", as opposed to other generation strategies that might be geometric or weighted in nature. Does that mean that Generator will end up growing more methods?

You should be able to create any distribution you want with the generator.