[Haskell-cafe] Stack & Test Coverage

[Hilco Wijbenga]

On Wed, Sep 4, 2019 at 10:14 PM Hilco Wijbenga <hilco.wijbenga at gmail.com> wrote:
>
> On Tue, Sep 3, 2019 at 11:14 PM Ben Gamari <ben at smart-cactus.org> wrote:
> >
> > On September 4, 2019 4:32:29 AM GMT+02:00, Hilco Wijbenga <hilco.wijbenga at gmail.com> wrote:
> > >Hi all,
> > >
> > >I have a very simple
> > >
> > >newtype Index = Index Int deriving (Eq, Ord, Enum)
> > >
> > >for which I have written tests with both HSpec and QuickCheck.
> > >Specifically, I have written tests for "==", "/=", "compare" (for all
> > >of EQ, LT, and GT), and "toEnum" & "fromEnum". So I thought I had
> > >_very_ thoroughly covered that one line. :-)
> > >
> > >Unfortunately, after running "stack clean ; stack test --coverage" I
> > >see in the HTML report that only Eq has been tested and that Ord and
> > >Enum have _no_ coverage (i.e. "never executed").
> > >
> > >(The module also has some regular functions and an explicit Show
> > >instance [all with tests] and those are apparently fine. So coverage
> > >does seem to work but not the way I expect?)
> > >
> > >What am I doing wrong?
> > >
> > >Cheers,
> > >Hilco
> > >_______________________________________________
> > >Haskell-Cafe mailing list
> > >To (un)subscribe, modify options or view archives go to:
> > >http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
> > >Only members subscribed via the mailman list are allowed to post.
> >
> > This sounds like it may be a ghc bug. Can you open a ticket at https://gitlab.haskell.org/ghc/ghc?
>
> Well, I wasn't expecting that. :-)
>
> I created a tiny test project with just the newtype and even removed
> Enum. I tried all LTSs listed at https://www.stackage.org from
> LST-3.22 all the way up to nightly-2019-09-04 and they all behave the
> same way. (Stack is _awesome_ for this.) I also tried "data" instead
> of "newtype". That made no difference.
>
> Are you sure this might be a GHC bug? Surely someone would have
> noticed sometime during the last 4 years? It would have been in GHC
> 7.10.2 already. Anyway, if "yes" then I'll create an issue. It should
> be very easy to reproduce. (I don't mean to doubt you but it's rather
> rare in software that it's "the compiler" and not "the developer". :-)
> I don't want to waste anyone's time.)

Mmm, so for sake of completeness:

----- Lib.hs
module Lib where
newtype Index = Index Int deriving (Eq, Ord, Show)

----- LibSpec.hs
module LibSpec (spec) where

import Test.Hspec
import Test.QuickCheck
import Lib

genValidIndexes :: Gen (Index, Index)
genValidIndexes = do lft <- choose (0, 100)
    rgt <- choose (0, 100)
    pure (Index lft, Index rgt)

prop_EqIndex :: (Index, Index) -> Bool
prop_EqIndex (lftIndex@(Index lft), rgtIndex@(Index rgt)) = (lftIndex
== rgtIndex) == (lft == rgt)

prop_OrdIndex :: (Index, Index) -> Bool
prop_OrdIndex (lftIndex@(Index lft), rgtIndex@(Index rgt)) = compare
lftIndex rgtIndex == compare lft rgt

spec :: Spec
spec
  = do
     describe "Index::Eq" $
       do it "Index 0 == Index 0" $ Index 0 == Index 0 `shouldBe` True
          it "Index 0 /= Index 1" $ Index 0 /= Index 1 `shouldBe` True
          it "Index 1 /= Index 0" $ Index 1 /= Index 0 `shouldBe` True
          it "prop_Eq" $ property $ forAll genValidIndexes prop_EqIndex
     describe "Index::Ord" $
       do it "compare (Index 0) (Index 0)" $ compare (Index 0) (Index
0) `shouldBe` EQ
          it "compare (Index 1) (Index 2)" $ compare (Index 1) (Index
2) `shouldBe` LT
          it "compare (Index 2) (Index 1)" $ compare (Index 2) (Index
1) `shouldBe` GT
          it "prop_OrdIndex" $ property $ forAll genValidIndexes prop_OrdIndex