Re: Type family equation violates injectivity?

i did some digging to see if theres a simple fix ... but then i realized that the current design they have for inlineR always assumes theres IO available in pure computations ... which they use the s parameter to thread into the pure vector code via reflection tricks so it looks like the "right" way fix this would be to have the MVector data type have both the state token for the the underlying resource AND whatever PrimMonad its running in, rather than smashing them together... especially since the ST monad and friend wont wont have the same state token as the underlying IO resource either way, what seems to be going on is that theres a mutex on which computations talk with the R RTS/instance the haskell code talks to, and theres a little bit of a mismatch in how they handle that .. either way, its not a ghc issue, its a design issue around vector and the api contract as the maintainers understand it vs how the package uses it. (though i repeat: i do think theres an issue in how they factor thestate token stuff ) On Sat, Dec 29, 2018 at 7:06 PM Carter Schonwald wrote:

To be clear : I’m annoyed with myself that this impacted a package that depends on vector, but this does seem to be the case that the newest bug fix release for vector actually revealed a broken design for the vector instances / data types in the inline-r package.

To;dr — I suggest patching inline-r to remove the s parameter in its immutable vector data types

On Sat, Dec 29, 2018 at 6:48 PM Carter Schonwald < carter.schonwald@gmail.com> wrote:

...

so i took a look .. (also the inline-r devs seem to have done a hackage revision so you wont hit that issue in your current setup if you do a cabal update ..) and it seems like the type definitions in inline-r are kinda bogus and you should get them patched ...

the MVector type class, and related type families, all assume your mutable type has the last two arguments as the io/state token and then the element type

eg basicLength :: v s a -> Int http://hackage.haskell.org/package/base-4.12.0.0/docs/Data-Int.html#t:Int

i looked at https://github.com/tweag/HaskellR/blob/1292c8a9562764d34ee4504b54d93248eb734... and

as a point of grounding this chat the injective type familly in question is defined by the follwoing

--#if MIN_VERSION_base(4,9,0)type family Mutable http://hackage.haskell.org/package/vector-0.12.0.2/docs/src/Data.Vector.Gene... (v http://hackage.haskell.org/package/vector-0.12.0.2/docs/src/Data.Vector.Gene... :: * -> *) = (mv http://hackage.haskell.org/package/vector-0.12.0.2/docs/src/Data.Vector.Gene... :: * -> * -> *) | mv -> v#elsetype family Mutable (v :: * -> *) :: * -> * -> *#endif

anyways, it looks like the Pure / immutable vector data type in inline-r has a spurious state token argument in its definition that shouldn't be there, OR there need to be two "s" params in inline-r instead of the one

heres the full code i linked to in question

-- | Mutable R vector. Represented in memory with the same header as 'SEXP'

-- nodes. The second type parameter is phantom, reflecting at the type level the -- tag of the vector when viewed as a 'SEXP'. The tag of the vector and the -- representation type are related via 'ElemRep'. data MVector s ty a = MVector { mvectorBase :: {-# UNPACK #-} !(SEXP s ty) , mvectorOffset :: {-# UNPACK #-} !Int32 , mvectorLength :: {-# UNPACK #-} !Int32 } -- | Internal wrapper type for reflection. First type parameter is the reified -- type to reflect. newtype W t ty s a = W { unW :: MVector s ty a } instance (Reifies t (AcquireIO s), VECTOR s ty a) => G.MVector (W t ty) a where

data Vector s (ty :: SEXPTYPE) a = Vector { vectorBase :: {-# UNPACK #-} !(ForeignSEXP ty) , vectorOffset :: {-# UNPACK #-} !Int32 , vectorLength :: {-# UNPACK #-} !Int32 }

type instance G.Mutable (W t ty s) = Mutable.W t ty Anyways, the fix here is to remove the s param from the Pure version of W and "Sexp Vector"

On Sat, Dec 29, 2018 at 6:16 PM Carter Schonwald < carter.schonwald@gmail.com> wrote:

...

were you using the same version of vector in both setups?

in the most recent vector release we made mutable type family injective in the vector package for ghc's that support it ...

On Sat, Dec 29, 2018 at 1:50 PM Dominick Samperi wrote:

...

When I use v8.6.3 of GHC under Ubuntu to install the inline-r package I get the error "Type family equation violates injectivity annotation," and a type variable on the LHS cannot be inferred from the RHS, due to the lack of injectivity (I suppose).

On the other hand, v8.0.2 of GHC (shipped with Haskell Platform under Ubuntu) does not have this problem (it has other problems).

Has something changed in the latest version of the compiler that might cause this? Possible work-around?

FYI, the line that triggers the error is: type instance G.Mutable (W t ty s) = Mutable.W t ty

The variable that cannot be inferred is 's'.

Thanks, Dominick _______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs