Don't use polysemy-plugin, remove IdempotentLoweringSpec

[KingoftheHomeless]

Stop-gap for https://github.com/polysemy-research/polysemy-zoo/issues/65. This does not address the underlying bug, so I won't close that issue. This wasn't as much work as I thought it would be.

[KingoftheHomeless]

@adamConnerSax Can you confirm if this works? I personally have issues trying to get 8.10 working with stack.

[adamConnerSax]

Looks like it. I had to do an "allow-newer: ghc-prim" for my case. Also, I wanted to do "allow-newer: random"--because the newer random is a major performance upgrade--but then I got a compilation error:

src/Polysemy/Random.hs:54:8: error:
    • Could not deduce: Polysemy.Internal.Union.LocateEffect
                          (Embed m0) r
                        ~ '()
        arising from a use of ‘embed’
      from the context: Member (Embed IO) r
        bound by the type signature for:
                   runRandomIO :: forall (r :: [(* -> *) -> * -> *]) a.
                                  Member (Embed IO) r =>
                                  Sem (Random : r) a -> Sem r a
        at src/Polysemy/Random.hs:52:1-68
      The type variable ‘m0’ is ambiguous
    • In a stmt of a 'do' block: q <- embed R.newStdGen
      In the expression:
        do q <- embed R.newStdGen
           snd <$> runRandom q m
      In an equation for ‘runRandomIO’:
          runRandomIO m
            = do q <- embed R.newStdGen
                 snd <$> runRandom q m
    • Relevant bindings include
        m :: Sem (Random : r) a (bound at src/Polysemy/Random.hs:53:13)
        runRandomIO :: Sem (Random : r) a -> Sem r a
          (bound at src/Polysemy/Random.hs:53:1)
   |
54 |   q <- embed R.newStdGen
   |        ^^^^^^^^^^^^^^^^^

[adamConnerSax]

So maybe, while you're there, it's worth bumping all the bounds, or at least all but random, though if that's easy to fix, it might be worth it...

[adamConnerSax]

Anyway, thanks for working on it!

[KingoftheHomeless]

Does that do it?

[KingoftheHomeless]

Forgot the tests. Those are significantly more hellish to rewrite, so I'm going with enabling the plugin for the tests only. Also, for some reason Travis now trips up on building IdempotentLoweringSpec, compared to in February. I have no idea why. Considering Idempotent Lowering has long since been de-facto deprecated in favour of Final, and ImpredicativeTypes has unspecified behaviour, I'll simply be removing that test, and not lose any sleep over it.

@adamConnerSax I'll merge this when you confirm this works.

[adamConnerSax]

I’ll test tomorrow AM!

[adamConnerSax]

All works on my end, thanks!