ghc-8.8.1-rc1 fails to build on s390x

Summary

ghc 8.8.1 RC1 build (using ghc-8.4.4) on s390x Linux fails while building ghctags.

Steps to reproduce

Build on Fedora s390x

Actual results

• https://kojipkgs.fedoraproject.org//work/tasks/1850/36451850/build.log (will be disappear after a couple of weeks)
• https://koji.fedoraproject.org/koji/taskinfo?taskID=36451850

BUILDSTDERR: utils/check-api-annotations/Main.hs:128:32: error:
BUILDSTDERR:     • GHC internal error: ‘GenericQ’ is not in scope during type checking, but it passed the renamer
BUILDSTDERR:       tcl_env of environment: [ahK2 :-> Type variable ‘r’ = r :: k0]
BUILDSTDERR:     • In the type signature:
BUILDSTDERR:         everything :: (r -> r -> r) -> GenericQ r -> GenericQ r
BUILDSTDERR:     |
BUILDSTDERR: 128 | everything :: (r -> r -> r) -> GenericQ r -> GenericQ r
BUILDSTDERR:     |                                ^^^^^^^^
BUILDSTDERR: make[1]: *** [utils/check-api-annotations/ghc.mk:18: utils/check-api-annotations/dist-install/build/Main.dyn_o] Error 1
BUILDSTDERR: make[1]: *** Waiting for unfinished jobs....
BUILDSTDERR: utils/ghctags/Main.hs:61:26: error:
BUILDSTDERR:     • GHC internal error: ‘FileName’ is not in scope during type checking, but it passed the renamer
BUILDSTDERR:       tcl_env of environment: [rF7M :-> ATcTyCon FileData :: *,
BUILDSTDERR:                                rF7N :-> APromotionErr RecDataConPE,
BUILDSTDERR:                                rF7O :-> ATcTyCon FoundThing :: *,
BUILDSTDERR:                                rF7P :-> APromotionErr RecDataConPE]
BUILDSTDERR:     • In the type ‘FileName’
BUILDSTDERR:       In the definition of data constructor ‘FileData’
BUILDSTDERR:       In the data declaration for ‘FileData’
BUILDSTDERR:    |
BUILDSTDERR: 61 | data FileData = FileData FileName [FoundThing] (Map Int String)
BUILDSTDERR:    |                          ^^^^^^^^
BUILDSTDERR: make[1]: *** [utils/ghctags/ghc.mk:18: utils/ghctags/dist-install/build/Main.dyn_o] Error 1
BUILDSTDERR: make: *** [Makefile:128: all] Error 2

Expected behavior

No build failure

Environment

• GHC version used: ghc-8.4.4

Optional:

• Operating System: Fedora Linux
• System Architecture: s390x

added Tbug ghc build failure needs triage s390x labels

changed the description

added compiler crash label

Note this is a regression (the earlier alphas built okay).
Not sure if this is related to the newer containers library in RC1.

Oh dear, this indeed looks nasty.

added Phighest label and removed needs triage label

@juhp, would you be able to trigger a build with another modified source distribution if I provided one? In particular I would like to try reverting the containers bump.

Sure I am very happy to test that.

Thanks!

I have uploaded a patched source distribution to http://home.smart-cactus.org/~ben/ghc/T16973.

Okay it is not containers...

https://koji.fedoraproject.org/koji/taskinfo?taskID=36486703

mentioned in issue #15411 (closed)

But wait I think I didn't apply the use USE_UNBOXED_ARRAYS patch again.

Yes with Disable-unboxed-arrays.patch it builds okay:

https://koji.fedoraproject.org/koji/taskinfo?taskID=36602852 (ignore the final packaging the error)

So reverting to containers-0.6.0.1 looks good to me.

So are you suggesting containers needs to skip unboxed arrays for this version? I'd rather lose that for a moment than hold years back from other containers changes. Better, of course, to find the GHC bug of course.

Should I try building without USE_UNBOXED_ARRAYS??

Yes. That flag only optimizes an operation in Data.Tree. You should be able to turn it off with no trouble. @bgamari please advise on a plan.

Okay I will try building 0.6.2.1 without USE_UNBOXED_ARRAYS - seems I was confused about this, somehow I thought the earlier patch was doing the opposite... Hopefully this still works then.

Okay sorry for false alarm and any stress caused :-)

container-0.6.2.1 and ghc-8.8.1-rc1 seem fine on unregisterized without USE_UNBOXED_ARRAYS.
https://koji.fedoraproject.org/koji/taskinfo?taskID=36608889

It might be good to make a general tracker issue for unregisterized ghc crashing with USE_UNBOXED_ARRAYS containers and close all the other related tickets as duplicates.

mentioned in issue #16998

marked this issue as related to #16998

It might be good to make a general tracker issue for unregisterized ghc crashing with USE_UNBOXED_ARRAYS containers and close all the other related tickets as duplicates.

Indeed; I had completely forgotten about #15411 (closed); it was only by luck that I stumbled upon it shortly after thinking of this ticket.

I have opened #16998 to track the root cause of this and will close this as a duplicate.

closed