inspection-testing-0.5.0.3/0.6 fails with GHC-9.12

[phadej]

<no location info>: error:
    panic! (the 'impossible' happened)
  GHC version 9.12.1:
	lookupIdSubst

But it's metadata explicltly says it works: ghc <9.13. Please fix either inspection-testing-0.5 or metadata.

[phadej]

Actually I see this error with inspection-testing-0.6 too. See well-typed/optics#512 for reproducer.

[nomeata]

Guess our test suite is incomplete here? @erikd you did the bump over at #80, do you feel like investigating?

[phadej]

In particular, there seems to be only on tests which causes the panic:

  , testCase "iover (imapped <% imapped) = iover (imapped % mapped)" $
    -- GHC 9.* applies a worker-wrapper transformation to the RHS.
    ghcGE90failure $(inspectTest $ 'lhs09 === 'rhs09)

in Optics.Tests.Core.

[phadej]

For the record, I was pointed to https://gitlab.haskell.org/ghc/ghc/-/issues/25468 but applying it fix to GHC-9.12.1-release didn't make the issue go away. (EDIT: looking at GHC issue tracker lookupIdSubst related issues are not uncommon).

It still might be a GHC issue, but I leave it to you to report upstream, as I don't know what inspection-testing is doing.

[phadej]

Another note: With -O0 the tests do compile, so it does look like an issue in GHC. That however doesn't help me, as there is no point to inspect -O0 code.

[erikd]

Guess our test suite is incomplete here? @erikd you did the bump over at #80, do you feel like investigating?

Yes, seems test suite is incomplete. I will see if I can extract the part of well-typed/optics#512 to add to the tests here.

[erikd]

I have reproduced the issue locally by cloning https://github.com/well-typed/optics/ and checking out the ghc-9.12 branch.

Reducing the file optics/tests/Optics/Tests/Core.hs to:

{-# LANGUAGE TemplateHaskell #-}
{-# OPTIONS_GHC -fplugin=Test.Inspection.Plugin -dsuppress-all #-}
module Optics.Tests.Core (coreTests) where

import Test.Tasty
import Test.Tasty.HUnit
import Test.Inspection

import Optics
import Optics.Tests.Utils

coreTests :: TestTree
coreTests = testGroup "Core"
  [ testCase "iover (imapped <% imapped) = iover (imapped % mapped)" $
    -- GHC 9.* applies a worker-wrapper transformation to the RHS.
    ghcGE90failure $(inspectTest $ 'lhs09 === 'rhs09)
  ]

-- Sometimes we need to eta expand, as without it pretty much equivalent code is
-- produced, but somewhat rearranged. Expanding allows us to get rid of these
-- differences to satisfy the check. However, when we do that we also need to
-- check whether the form that is not eta-expanded optimizes away internal
-- representation correctly.

lhs09, rhs09
  :: (FunctorWithIndex i t, FunctorWithIndex j s)
  => (i -> a -> b)
  -> t (s a)
  -> t (s b)
lhs09 = iover (imapped <% imapped)
rhs09 = iover (imapped % mapped)

is sufficient to trigger the GHC bug.

[erikd]

Raised a GHC issue: https://gitlab.haskell.org/ghc/ghc/-/issues/25631

Now trying to extract the code that triggers this bug for inclusion in the test suite.

[erikd]

When I cargo culted minimal parts of inpection-testing into the file optics/tests/Optics/Tests/Core.hs the GHC panic was not triggered. 😭

[erikd]

SPJ' latest comment: https://gitlab.haskell.org/ghc/ghc/-/issues/25631#note_602995

I had a look at the code but could make very little sense of it. @nomeata ?

[Mikolaj]

FYI, I've just used inspection-testing with a couple weeks old GHC HEAD with great success and no crashes. But I haven't attempted reproducing your problem.