Adjust for new additions in upcoming random-1.3

[lehins]

Hey @Shimuuar I am working on some changes in the random-1.3 stateful interface in haskell/random#153 Could you take a look and see if you find it sensible.

Addition of modifyGen is more useful for stateful generators that are regular wrappers around RandomGen rather than for the ones like mwc implements. It also allows for a mutable Gen to be overwritten with the Seed using overwriteGen

[hsyl20]

random-1.3 has been published. It makes the currently released mwc-random fail to build. Could you make a Hackage package revision to exclude random > 1.3 until this PR is merged and released?

[Shimuuar]

@hsyl20 I've updated bound on hackage

@lehins What are your plans on SeedGen?

[Shimuuar]

I think, I'll need to add both random 1.2 and random 1.3 to build matrix

[lehins]

@lehins What are your plans on SeedGen?

@Shimuuar Very good question.

There are two approaches that can be taken. One is implemented in #96 That approach is efficient, but it would not work for seed being compatible between big-endian and littel-endian machines and it requires a newer version of vector.

There is an alternative version I am planning to submit a PR for as well, that will be simpler and will work just as well regardless of endianness, but it will be slowe, since it will have to go through a list.

I was really waiting for this PR to be merged and random-1.3 to be released before I had both of the solutions implemented for the SeedGen instance.

[lehins]

FYI @Shimuuar I am working on fixing up this PR to pass CI

[Shimuuar]

RE SeedGen not every vector of 258 Word32 is valid generator last two words have special meaning. One at ioff=256 is index and must be in [0..255] range. And one at coff=257 is carry at must be smaller than aa=1540315826. initialize does normaliztion

[lehins]

RE SeedGen not every vector of 258 Word32 is valid generator last two words have special meaning. One at ioff=256 is index and must be in [0..255] range. And one at coff=257 is carry at must be smaller than aa=1540315826. initialize does normaliztion

Yeah, I've seen that. We can do normalization by the means of runST . save . initialize or something along those lines in the implementation System.Random.toSeed function.

I only implemented that instance so far as a proof of concept, just to make sure the new SeedGen interface in random is viable.

[lehins]

Looks like there is also a need for a lower bound revision on hackage for tasty-quickcheck >= 0.10.2.

[lehins]

I think this PR is ready to go.

[Shimuuar]

I think PR is good to go. @lehins thanks! We still need to test both random-1.2 and 1.3 on CI. But that could be done separatelu