Arbitrary generation

[jkarni]

Currently we use Arbitrary for everything. This runs into an issue that makes testing particular endpoints not very effective. To see why, consider a typical CRUD set of endpoints (where e.g. "user/:id" is one of the GET endpoints). With a clean DB, arbitrary generation of IDs will almost never result in a query to an existing user - even if servant-quickcheck POSTs a new arbitrary user, the likelihood of it GETting that user is very small.

I think the way to go is to have a class with an IO a (rather than Gen a) method. This would allow library users to for example specify that Id should be taken from the list of IDs in the database.

[soenkehahn]

I haven't looked into it a lot, but maybe this contains some relevant ideas:
http://quviq.com/documentation/eqc/eqc_statem.html

(From http://quviq.com/documentation/eqc/index.html)

[jkarni]

Yeah, this would be nice. I worry adding this capability on top of (Haskell) QuickCheck would be a lot more work that just having IO a, though.