Doesn't work with data-fix-0.0.7

[mdorman]

To be clear, I do not in any way, shape or form understand this failure.

With version 0.0.7 of the data-fix package, hnix fails to build its tests with the following error:

tests/ParserTests.hs:340:23: error:
    • No instance for (Data.Functor.Classes.Eq1 NExprF)
        arising from a use of ‘assertEqual’
    • In the expression:
        assertEqual ("Parsing data file " ++ file) expected actual
      In a case alternative:
          Success actual
            -> assertEqual ("Parsing data file " ++ file) expected actual
      In a stmt of a 'do' block:
        case res of {
          Success actual
            -> assertEqual ("Parsing data file " ++ file) expected actual
          Failure err
            -> assertFailure
               $ "Unexpected error parsing data file `"
                 ++ file ++ "':\n" ++ show err }

On the other hand, with version 0.0.4, it builds fine.

The part that I find baffling is that the only difference between those versions would appear to be:

diff -uNr /home/mdorman/t/data-fix-0.0.4/ /home/mdorman/t/data-fix-0.0.7/
diff -uNr /home/mdorman/t/data-fix-0.0.4/data-fix.cabal /home/mdorman/t/data-fix-0.0.7/data-fix.cabal
--- /home/mdorman/t/data-fix-0.0.4/data-fix.cabal	2017-03-29 05:37:43.000000000 -0400
+++ /home/mdorman/t/data-fix-0.0.7/data-fix.cabal	2017-09-03 18:04:27.000000000 -0400
@@ -1,6 +1,6 @@
 Name:            data-fix
-Version:         0.0.4
-Cabal-Version:   >= 1.6
+Version:         0.0.7
+Cabal-Version:   >= 1.10
 License:         BSD3
 License-file:    LICENSE
 Author:          Anton Kholomiov
@@ -12,7 +12,7 @@
   Fixpoint types and recursion schemes. If you define your AST as
   fixpoint type, you get fold and unfold operations for free.
   .
-  Thanks for contribution to: Matej Kollar
+  Thanks for contribution to: Matej Kollar, Herbert Valerio Riedel
 
 Stability:       Experimental
 
@@ -24,7 +24,8 @@
     Location: https://github.com/anton-k/data-fix
 
 Library
-  Build-depends: base >= 4, base < 5
+  Default-Language: Haskell2010
+  Build-depends: base >= 4.7, base < 5
   Hs-source-dirs: src/
 
   ghc-options: -Wall

Diff finished.  Wed Sep 13 06:24:32 2017

[mdorman]

Just as a note, I'm building on nixos, synced to roughly the latest commit. The command I used to run the build was nix-build --no-build-output --attr haskellPackages.hnix ~/src/nixpkgs. When I started getting the error, I bisected to the change that updated data-fix, backed out that one single update to the haskellPackages set, and it built again.

[Profpatsch]

Here’s the whole change:

Probably something in base or tasty-hunit changed.

[Profpatsch]

Oh, @jwiegley, 0.3.4 was uploaded a few days ago, but the last commit on github is from May?! (Curse hackage being unaware of the source code repo.)

[anton-k]

Hi! i'm the author of data-fix. Unfortunately it doesn't compile with the latest ghc. There are some incompatible changes regarding instances of Eq vs Eq1 and Ord vs Ord1. I don't completely understand them right now, but if my package causing any trouble I'm willing to learn and cooperate :)

[anton-k]

There was a PR that proposed the updates. I have accpeted it. It's data-fix-0.6 but then I've reverted it in 0.7 since some of my libraries doesn't work well with it. Maybe I need to update the libraries also and keep the change to make your packages compile!

Can you try out the data-fix-0.0.6. Does it work for you?

[anton-k]

The 0.0.7 is exactly as 0.0.4

[mdorman]

@Profpatsch It's not one of the deps, because I can add a data-fix_0_0_4 expression, point hnix to that expression and it works---so that's definitely the source of the issue.

@anton-k I don't quite understand the nature of the failure myself---I hope to do a little investigation, but I'm hardly an expert. I do wonder if the change in the haskell version is actually the issue?

[Profpatsch]

With 0.0.6, the error is the same:

→ ghc-pkg list
/nix/store/0k3j1cg179shkdnsk86p01cvkaxl42ap-ghc-8.0.2-with-packages/lib/ghc-8.0.2/package.conf.d
    Cabal-1.24.2.0
    …
    data-fix-0.0.6
…

→ cabal test
Preprocessing library for hnix-0.3.4..
Building library for hnix-0.3.4..
…
[3 of 5] Compiling ParserTests      ( tests/ParserTests.hs, dist/build/hnix-tests/hnix-tests-tmp/ParserTests.o )

tests/ParserTests.hs:348:23: error:
    • No instance for (Data.Functor.Classes.Eq1 NExprF)
        arising from a use of ‘assertEqual’
    • In the expression:
        assertEqual ("Parsing data file " ++ file) expected actual
      In a case alternative:
          Success actual
            -> assertEqual ("Parsing data file " ++ file) expected actual
      In a stmt of a 'do' block:
        case res of {
          Success actual
            -> assertEqual ("Parsing data file " ++ file) expected actual
          Failure err
            -> assertFailure
               $ "Unexpected error parsing data file `"
                 ++ file ++ "':\n" ++ show err }

The corresponding failing function is:

assertParseFile :: FilePath -> NExpr -> Assertion
assertParseFile file expected = do
  res <- parseNixFile $ "data/" ++ file
  case res of
    Success actual -> {- HERE -} assertEqual ("Parsing data file " ++ file) expected actual
    Failure err    -> assertFailure $ "Unexpected error parsing data file `" ++ file ++ "':\n" ++ show err

I’m on the latest nixpkgs master with these haskell packages. hnix is checked out from the github master.

[Profpatsch]

Update: can confirm that it works with 0.0.4 and hnix 0.3.4. But not with 0.0.6 or 0.0.7.

[jwiegley]

@Profpatsch Do you recommend setting an upper bound, or does it now work with 0.0.7?

[Profpatsch]

Yes, it works with 0.0.7. So a lower bound on 0.0.7 might be sensible, or an exclusion of 0.0.6.

Based on my input, the author bumped the version to 0.2.0, so they can indicate non-API-breaking changes.

[jwiegley]

OK, I've changed the latest version of hnix to depend on at least 0.0.7 of data-fix.