Feat/integrated shrinking

[sir4ur0n]

This is a reopening with additional fixes to original PR #109 adding QCheck2 and Integrated shrinking.

[sir4ur0n]

I think I have managed to make function shrinking work again 🎉 At least the first tests are positive, see:

let test_basic : QCheck2.Test.t =
  let open QCheck2 in
  Test.make (fun1 Observable.int bool) (fun (Fun (_, f)) -> f 42 = f 17)

test_function alias test/runtest (exit 1)
(cd _build/default/test && ./test_function.exe)
random seed: 52241607
randomgenerated error fail pass / total     time test name
[✗]    1    0    1    0 /  100     0.0s anon_test_1

--- Failure --------------------------------------------------------------------

Test anon_test_1 failed (2 shrink steps):

{17 -> true; _ -> false}
================================================================================
failure (1 tests failed, 0 tests errored, ran 1 tests)

I am going to test it further today, add some documentation on all its APIs, then we should be good to go!

[sir4ur0n]

@c-cube IMHO this PR is done.

In the last days I have implemented a first version of function shrinking. It's not perfect but it works (ish).
I also documented all function-related types, modules and functions, e.g. Fn or Tuple.

Please do another (hopefully last 😁 ) review

[sir4ur0n]

I am currently migrating all existing Tezos tests to QCheck2 as a dogfooding experiment.

Current status:

• I found some missing bits in the migration guide, added it
• I noticed there was no getter for collect and stats, added it
• I notice a big performance regression on Gen.string, currently investigating

I think the bulk of the PR can still be reviewed (but this should not be merged until such obvious regressions are fixed 😁 )

[sir4ur0n]

I just pushed:

• 6720de8 significantly improves performance of the char generator, which in turn improves perf of string generators
• e44b8ca puts more laziness everywhere related to shrinking. Some measurements tend to show a ~40% perf improvement on tests (that don't fail/shrink, which is the majority of time)

I'd be happy to get a review on these recent code changes as I am not experienced in OCaml performance 🙏

I'm resuming migrating my project to QCheck2 to see if I detect other problems

[sir4ur0n]

Tezos QCheck tests all pass with QCheck2.

I note that the total run time is slower with QCheck2 (~60s with QCheck2, ~50s with QCheck1).
I don't find this outraging considering the added feature of auto shrinking, but if anyone wants to take a look to see if more performance can be squeezed, please do so 😁

[c-cube]

Very nice. I think it's almost there, I just added a bunch of suggestions, but I think this can be merged real soon :)

[sir4ur0n]

@c-cube I believe I have managed all threads, please review again 🙏

[c-cube]

But that's not rétro-compatible, is it? That's my point

[c-cube]

Alright team, this time I think it's merged for good! 🎉

Thank you very much @sir4ur0n for your hard work and patience. This was quite the epic PR!