FunctorWithIndex.imap (or Eq) in version 0.6.2.1 is broken

[blackgnezdo]

This test passes in GHC 8.6.4 and fails in 8.10.1 with lens-4.19.2:

import Control.Lens
import GHC.Exts (IsList(..))
import qualified Data.IntMap as IM

main :: IO ()
main = do
    let f i a = 5 + i * a
        g i a = a + 2 * i
        m = fromList [(-2,1),(0,-4)] :: IM.IntMap Int
        -- This is a FunctorWithIndex law.
        x = imap f (imap g m)
        y = imap (\i -> f i . g i) m
    print m
    print x
    print y
    print $ x == y

8.6 prints the expected:

fromList [(-2,1),(0,-4)]
fromList [(-2,11),(0,5)]
fromList [(-2,11),(0,5)]
True

8.10 prints:

fromList [(-2,1),(0,-4)]
fromList [(-2,11),(0,5)]
fromList [(0,5),(-2,11)]
False

[blackgnezdo]

@RyanGlScott FYI, in case the problem is in lens.

[sjakobi]

Thanks for the report, @blackgnezdo. Could you reduce the reproducer, so it depends only on containers? It might be worth checking it with master – there have been quite a few fixes of IntMap traversals since the last release: https://github.com/haskell/containers/blob/master/containers/changelog.md

[treeowl]

Yeah, this smells like that.

[blackgnezdo]

Sadly I can't get a standalone repro outside of our ghc environment at all. Let me get at least this much isolated before other people spend time on it.

[sjakobi]

I've had a brief look at imap in lens. AFAICT the definition is derived from the TraversableWithIndex instance, which is defined with traverseWithKey:

http://hackage.haskell.org/package/lens-4.19.2/docs/src/Control.Lens.Indexed.html#line-703

So I'm quite certain that you're hitting the bug that we've already fixed on master:

Fix traverse and traverseWithKey for IntMap, which would
previously produce invalid IntMaps when the input contained
negative keys

@treeowl BTW, when do you expect to cut the next release? With the possible exception of #729, everything seems to be ready.

[blackgnezdo]

Yes, I verified 7c40db1 fixes the problem.