Needs to be updated to account for remote-json's use of natural-transformation

[ecaustin]

This remote-json commit broke compilation for hermit-shell:
ku-fpg/remote-json@ed3597c

There may be other bugs past this one, but for sure the definition of the remote session needs to be updated:

session :: JSONRPC.Session
session = JSONRPC.session
        $ (if debug 
           then traceSessionAPI "HERMIT-remote-json"
           else id)
        $ sendr

The use of id doesn't properly maintain the scope of the quantified variable introduced by the natural transformation type (~>).

I'm assuming the appropriate fix here is to add a natural identity transformation to natural-transformation and make hermit-shell's reliance on that library explicit?

[andygill]

Nice find. What (GHC) version fails?

[ecaustin]

It for sure fails with GHC 7.10.3 and 7.10.2.

I can draw up a smaller example to test with GHC 7.8.x since HERMIT doesn't support that version of GHC anymore.

The only reason it would not fail would be a change to the typechecker between versions that broke the semantics of identity, right?

That'd be bad...

[andygill]

I'm actively heading towards remote-json, which is going to use the remote-monad package. And I'm happy for you to make any changes need to remote-json, or direct me to do so. But the code works in 7.10.3 for me. I'll add a travis instance of 8.X, and see what happens.

[ecaustin]

Do you have a local commit you haven't pushed, or are you using an older version of any of the relevant libraries?

Here's the exact error I get:

[20 of 55] Compiling HERMIT.GHCI.Client ( src/HERMIT/GHCI/Client.hs, dist/build/HERMIT/GHCI/Client.o )

src/HERMIT/GHCI/Client.hs:34:17:
    Couldn't match type ‘forall x1. SessionAPI x1 -> IO x1’
                   with ‘SessionAPI x -> IO x’
    Expected type: SessionAPI
                   natural-transformation-0.3:Control.Natural.~> IO
                   -> SessionAPI x -> IO x
      Actual type: SessionAPI
                   natural-transformation-0.3:Control.Natural.~> IO
                   -> SessionAPI natural-transformation-0.3:Control.Natural.~> IO
    In the expression: id
    In the expression:
      (if debug then traceSessionAPI "HERMIT-remote-json" else id)

That's with a fresh checkout of natural-transformation, remote-json, and hermit-shell and GHC 7.10.3. I can include my entire package list if you think the issue might be with an old version of one of the compat libraries or something.

[andygill]

No local changes. A local build of only remote-json though. Yes, what versions of the packages are you using.

[ecaustin]

/home/e/setup/remote-json/.cabal-sandbox/x86_64-linux-ghc-7.10.3-packages.conf.d
   aeson-0.9.0.1
   attoparsec-0.13.0.1
   data-default-class-0.0.1
   dlist-0.7.1.2
   exceptions-0.8.1
   hashable-1.2.4.0
   mtl-2.2.1
   natural-transformation-0.2
   primitive-0.6.1.0
   remote-json-0.2
   scientific-0.3.4.4
   stm-2.4.4.1
   syb-0.6
   text-1.2.2.0
   transformers-compat-0.4.0.4
   unordered-containers-0.2.5.1
   vector-0.11.0.0

There's my package list dump.
This VM is less than a day old so everything should be the most recent version.

I can confirm I'm still having the issue with a local build of remote-json and using natural-transformation from Hackage.

[andygill]

Can you try build just remote-json, in a local sandbox?

[ecaustin]

Sorry, I misunderstood what you were asking for. I've updated my previous comment.

[ecaustin]

I've identified a simpler test case that only requires natural-transformation:
ku-fpg/natural-transformation#9

[andygill]

I've reproduced the error (on lilybank) and working on it.

src/HERMIT/GHCI/Client.hs:34:17:
    Couldn't match type ‘forall x1. SessionAPI x1 -> IO x1’
                   with ‘SessionAPI x -> IO x’
    Expected type: SessionAPI
                   natural-transformation-0.3:Control.Natural.~> IO
                   -> SessionAPI x -> IO x
      Actual type: SessionAPI
                   natural-transformation-0.3:Control.Natural.~> IO
                   -> SessionAPI natural-transformation-0.3:Control.Natural.~> IO
    In the expression: id
    In the expression:
      (if debug then traceSessionAPI "HERMIT-remote-json" else id)

[andygill]

We might be able to avoid this if we use

Nat f g -> Nat f g

rather than

(forall a . f a -> g a) -> (forall a . f a -> g a)