Looking for people who are using Hypothesis and are willing to say so

[David MacIver]

Actually one of the things that's helped the most in the course of
designing Hypothesis is the realisation that types are something of a red
herring for this sort of testing. Thinking purely in terms of custom
generators helps a lot, because it means you can do things like specify
size bounds on lists, integers, etc. as well as map and filter over the
resulting data (e.g. lists(integers().map(lambda x: x * 2), min_size=1,
max_size=10). Lists of length between 1 and 10 only containing even
integers. So in this regard the design of Hypothesis should be considered
more closely related to that of the Erlang than the Haskell quickcheck
(although I'd only ever used statically typed quickchecks before writing
Hypothesis).

In particular Hypothesis's strategies are best thought of in terms of how
to provide data as the argument rather than a type - you can't check
whether a given value is producable from a given strategy.

I was originally thinking it would be worth writing some auto derivation
functionality for using the new python 3 type annotations, but I actually
don't think it would be useful. You lose far too much flexibility.



On 24 June 2015 at 23:13, Paul Rubin <no.email at nospam.invalid> wrote:

> David MacIver <david at drmaciver.com> writes:
> > Author of Hypothesis here. (If you don't know what Hypothesis is, you're
> > probably not the target audience for this email but you should totally
> > check it out: https://hypothesis.readthedocs.org/
>
> Oh very cool: a QuickCheck-like unit test library.  I heard of something
> like that for Python recently, that might or might not have been
> Hypothesis.  I certainly plan to try it out.  The original QuickCheck
> (for Haskell) used the static type signatures on the functions under
> test to know what test cases to generate, but Erlang QuickCheck has had
> some good successes, including finding some subtle bugs during
> development in the HAMT (Clojure-like hash array mapped trie)
> implementation just released with Erlang/OTP 18.0 this week.
>
> I see Hypothesis use decorators that look sort of like Erlang Dialyzer
> so that can help with test cases.  Maybe later, it use Python 3 type
> annotations, though I think those are still much less precise than
> Dialyzer or Haskell types.
> --
> https://mail.python.org/mailman/listinfo/python-list
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/python-list/attachments/20150625/17c97221/attachment.html>