Part of this PR is meant to address #241, which was an unhappy reaction to how rand was evolving at the time. Since then, rand has slowed down its development and slimmed its dependency tree. It's still not quite as small as what quickcheck needs (which is really just a non-crypto PRNG), but it's closer.
One idea I and @matklad had was to just remove rand from the public API of QuickCheck. Doing this has a cost: it removes the ability to provide your own RNG to quickcheck and instead moves the use of the RNG to an implementation detail. My feeling is that providing one's own RNG to quickcheck was seldom-if-ever used, so I judged this to be an acceptable cost. In return, we are no longer pinned to rand's release cycle and we now have the freedom to switch out the RNG in the future if we want. For now, we upgrade to rand 0.8 and otherwise leave it alone for now.
In addition to losing the ability to provide your own RNG, we also lose a lot of the lower level RNG convenience routines provided by rand. I believe we can capture much of that by just exposing concrete routines on quickcheck::Gen, in addition to the various Arbitrary impls.
The reasons why the Gen trait existed in the first place were:
I believe I copied the design as faithfully as possible from Haskell's QuickCheck, and IIRC, it is generic over the RNG.
At the time I wrote QuickCheck, RNGs were part of the standard library. Once RNGs moved out into the crate ecosystem, the costs of making that crate a public dependency manifested itself.
Anywho, I've written all this out so that it's clear what has changed and why. In particular, if you find yourself at this PR because some functionality in quickcheck has been removed, then please open a new issue that includes the problem you're trying to solve and we'll try to figure out how to move forward.
Part of this PR is meant to address #241, which was an unhappy reaction to how
randwas evolving at the time. Since then,randhas slowed down its development and slimmed its dependency tree. It's still not quite as small as what quickcheck needs (which is really just a non-crypto PRNG), but it's closer.One idea I and @matklad had was to just remove
randfrom the public API of QuickCheck. Doing this has a cost: it removes the ability to provide your own RNG to quickcheck and instead moves the use of the RNG to an implementation detail. My feeling is that providing one's own RNG to quickcheck was seldom-if-ever used, so I judged this to be an acceptable cost. In return, we are no longer pinned torand's release cycle and we now have the freedom to switch out the RNG in the future if we want. For now, we upgrade torand 0.8and otherwise leave it alone for now.In addition to losing the ability to provide your own RNG, we also lose a lot of the lower level RNG convenience routines provided by
rand. I believe we can capture much of that by just exposing concrete routines onquickcheck::Gen, in addition to the variousArbitraryimpls.The reasons why the
Gentrait existed in the first place were:Anywho, I've written all this out so that it's clear what has changed and why. In particular, if you find yourself at this PR because some functionality in quickcheck has been removed, then please open a new issue that includes the problem you're trying to solve and we'll try to figure out how to move forward.