Build failure on PowerPC 64-bit big endian

Building GHC 8.6.1 alpha 2 fails with these errors on Powerpc 64-bit big endian Linux:

[ 5076s] utils/check-ppr/Main.hs:44:18: error:
[ 5076s]     * GHC internal error: `astFile' is not in scope during type checking, but it passed the renamer
[ 5076s]       tcl_env of environment: [agVj :-> Identifier[libdir::FilePath, NotLetBound],
[ 5076s]                                agVk :-> Identifier[fileName::String, NotLetBound],
[ 5076s]                                agVl :-> Identifier[p::a, NotLetBound],
[ 5076s]                                agVo :-> Identifier[anns::ApiAnns, TopLevelLet [agVl :-> p] False],
[ 5076s]                                agVp :-> Identifier[pragmas::String, TopLevelLet [agVo :-> anns] False],
[ 5076s]                                agVq :-> Identifier[newFile::FilePath, TopLevelLet [agVk :-> fileName] False],
[ 5076s]                                rfRp :-> Identifier[usage::String, TopLevelLet [] True],
[ 5076s]                                rfU1 :-> Identifier[main::IO (), TopLevelLet [] True],
[ 5076s]                                rfU2 :-> Identifier[testOneFile::FilePath
[ 5076s]                                                                 -> String
[ 5076s]                                                                 -> IO (), TopLevelLet [] True],
[ 5076s]                                rfU3 :-> Identifier[parseOneFile::FilePath
[ 5076s]                                                                  -> FilePath
[ 5076s]                                                                  -> IO
[ 5076s]                                                                       ParsedModule, TopLevelLet [] True],
[ 5076s]                                rfU4 :-> Identifier[getPragmas::ApiAnns -> String, TopLevelLet],
[ 5076s]                                rfU5 :-> Identifier[pp::forall a.
[ 5076s]                                                        Outputable a =>
[ 5076s]                                                        a -> String, TopLevelLet [] True]]
[ 5076s]     * In the first argument of `writeFile', namely `astFile'
[ 5076s]       In a stmt of a 'do' block: writeFile astFile origAst
[ 5076s]       In the expression:
[ 5076s]         do p <- parseOneFile libdir fileName
[ 5076s]            let pped = pragmas ++ "\n" ++ pp (pm_parsed_source p)
[ 5076s]                anns = pm_annotations p
[ 5076s]                ....
[ 5076s]            writeFile astFile origAst
[ 5076s]            writeFile newFile pped
[ 5076s]            ....
[ 5076s]    |
[ 5076s] 44 |        writeFile astFile origAst
[ 5076s]    |                  ^^^^^^^
[ 5076s] utils/check-ppr/ghc.mk:18: recipe for target 'utils/check-ppr/dist-install/build/Main.o' failed
[ 5076s] make[1]: *** [utils/check-ppr/dist-install/build/Main.o] Error 1
[ 5076s] make[1]: *** Waiting for unfinished jobs....
[ 5077s] 
[ 5077s] utils/check-api-annotations/Main.hs:116:32: error:
[ 5077s]     * GHC internal error: `GenericQ' is not in scope during type checking, but it passed the renamer
[ 5077s]       tcl_env of environment: [agZx :-> Type variable `r' = r :: k0]
[ 5077s]     * In the type signature:
[ 5077s]         everything :: (r -> r -> r) -> GenericQ r -> GenericQ r
[ 5077s]     |
[ 5077s] 116 | everything :: (r -> r -> r) -> GenericQ r -> GenericQ r
[ 5077s]     |                                ^^^^^^^^
[ 5077s] utils/check-api-annotations/ghc.mk:18: recipe for target 'utils/check-api-annotations/dist-install/build/Main.o' failed
[ 5077s] make[1]: *** [utils/check-api-annotations/dist-install/build/Main.o] Error 1
[ 5077s] 
[ 5077s] utils/ghctags/Main.hs:61:36: error:
[ 5077s]     * GHC internal error: `FoundThing' is not in scope during type checking, but it passed the renamer
[ 5077s]       tcl_env of environment: [rCTr :-> ATcTyCon FileData :: *,
[ 5077s]                                rCTs :-> APromotionErr RecDataConPE,
[ 5077s]                                rCTw :-> ATcTyCon FileName :: *]
[ 5077s]     * In the type `[FoundThing]'
[ 5077s]       In the definition of data constructor `FileData'
[ 5077s]       In the data declaration for `FileData'
[ 5077s]    |
[ 5077s] 61 | data FileData = FileData FileName [FoundThing] (Map Int String)
[ 5077s]    |                                    ^^^^^^^^^^
[ 5077s] utils/ghctags/ghc.mk:18: recipe for target 'utils/ghctags/dist-install/build/Main.o' failed

Alpha 2 builds fine on PowerPC 64-bit little endian Linux.

I am setting version to 8.5 as there is no tag for 8.6.1 alpha 2 yet.

Trac metadata

Trac field	Value
Version	8.5
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture

changed weight to 5

added Tbug Trac import labels

mentioned in issue #15411 (closed)

Currently my last known good commit is changeset:d78dde9f.

At changeset:bdfc85b8 I see the following tests failures during validate:

Unexpected results from:
TEST="ManyAlternatives ManyConstructors T10520 T10858 T12007 T12425 T12545 T12707 T13035 T13379 T13615 T13825-ghci T14894 T1969 T2182ghci T3064 T3294 T4801 T5321FD T5321Fun T5435_v_asm_b T5631 T6048 T7253 T7386 T783 T8557 T9233 duplicaterecfldsghci01 ghc-e006 ghci.prog007 ghci006 ghci039 ghci045 ghci049 ghci052 ghci053 ghcirun001 ghcirun002 heap_all print002 print003 print006 print007 print008 print010 print012 print013 print014 print018 print019 print021 print025 print031 print034"

SUMMARY for test run started at Wed Jul 18 12:35:24 2018 CEST
 1:18:37 spent to go through
    6386 total tests, which gave rise to
   19686 test cases, of which
   13227 were skipped

      38 had missing libraries
    6190 expected passes
     176 expected failures

       2 caused framework failures
       1 caused framework warnings

Note the large number of GHCi tests failing.

I will now bisect in that range above.

The breaking commit is changeset:ec22f7dd. I have found an endianness issue with this commit in compiler/ghci/RtClosureInspect.hs.

I am just validating a patch on ppc64.

My patch was not quite good enough.

Here is another endianness issue:

        BCO -> do
// removed checks
            let splitWord = rawWds !! 3
            pure $ BCOClosure itbl (pts !! 0) (pts !! 1) (pts !! 2) 
// on big endian this is the wrong way around
                (fromIntegral splitWord)
                (fromIntegral $ shiftR splitWord (wORD_SIZE_IN_BITS `div` 2))
                (drop 4 rawWds)

I need a way to tell whether I am running on a big endian system. Is there any way to know inside without access to DynFlags? Is there even a portable way to read the first half-word from a word in memory?

Actually, ticket:15399#comment:157204 applies to AP and PAP closures as well.

Here is a fix for one of the 36 failing GHCi tests. It fixes three endianness issues introduced by changeset:ec22f7dd.

Trac metadata

Trac field	Value
Differential revisions	- → D5001

assigned to @trommler

mentioned in commit d7cb1bbc

changed milestone to %8.6.1

added 1 deleted label

closed

Merged to ghc-8.6 in 3795b454.

Trac metadata

Trac field	Value
Resolution	Unresolved → ResolvedFixed

removed 1 deleted label

mentioned in issue #15916 (closed)

reopened

The issue shown in the ticket is a duplicate of #15411 (closed).

In the comments for this ticket we found the first breaking commit, which is unrelated to the issue on this ticket, and fixed (Phab:5001) one (out of 36) of the broken tests (see ticket:15399#comment:157246).

I reopen the ticket so I won't forget to create a new ticket for the remaining 35 failing tests.

Trac metadata

Trac field	Value
Resolution	ResolvedFixed → Unresolved
Related	- → #15411 (closed)

I will submit a follow-up ticket.

added ghc build failure label

The above reported tests do not fail anymore in HEAD with the workaround from #15411 (closed) applied.

closed

added Pnormal label