Marge Bot pushed to branch master at Glasgow Haskell Compiler / GHC

Commits:

4 changed files:

Changes:

  • changelog.d/T27314.md
    1
    +section: compiler
    
    2
    +issues: #27314
    
    3
    +mrs: !16118
    
    4
    +synopsis:
    
    5
    +  Fix spurious ``-Wincomplete-uni-patterns`` warning under ``-finfo-table-map``.
    
    6
    +description:
    
    7
    +  The pattern-match checker now ignores ticks when comparing scrutinees in
    
    8
    +  its CoreMap, so long-distance information is no longer lost across
    
    9
    +  function-application scrutinees because debug source annotations
    
    10
    +  (e.g. SourceNotes added by ``-finfo-table-map``) were inserted.

  • compiler/GHC/HsToCore/Pmc/Solver.hs
    ... ... @@ -1000,8 +1000,9 @@ makeDictsCoherent (Case scrut bndr ty alts)
    1000 1000
           , let expr' = makeDictsCoherent expr ]
    
    1001 1001
     makeDictsCoherent (Cast expr co)
    
    1002 1002
       = Cast (makeDictsCoherent expr) co
    
    1003
    -makeDictsCoherent (Tick tick expr)
    
    1004
    -  = Tick tick (makeDictsCoherent expr)
    
    1003
    +makeDictsCoherent (Tick _tick expr)
    
    1004
    +  -- See Wrinkle (UD1) in Note [Unique dictionaries in the TmOracle CoreMap]
    
    1005
    +  = makeDictsCoherent expr
    
    1005 1006
     makeDictsCoherent ty@(Type {})
    
    1006 1007
       = ty
    
    1007 1008
     makeDictsCoherent co@(Coercion {})
    
    ... ... @@ -1061,6 +1062,25 @@ In the end, replacing dictionaries with an error value in the pattern-match
    1061 1062
     checker was the most self-contained, although we might want to revisit once
    
    1062 1063
     we implement a more robust approach to computing equality in the pattern-match
    
    1063 1064
     checker (see #19272).
    
    1065
    +
    
    1066
    +Wrinkle (UD1): ticks
    
    1067
    +--------------------
    
    1068
    +'makeDictsCoherent' also drops all ticks. The CoreMap key represents
    
    1069
    +value-level equality, which ticks never affect.
    
    1070
    +
    
    1071
    +Example (#27314): with -finfo-table-map every record-selector use site is
    
    1072
    +wrapped in a 'SourceNote' carrying that site's span (see
    
    1073
    +Note [Record-selector ticks] in GHC.HsToCore.Ticks). Given
    
    1074
    +
    
    1075
    +    data Box = Box { unBox :: Maybe Int }
    
    1076
    +    f b = case unBox b of
    
    1077
    +      Nothing -> 0
    
    1078
    +      Just _  -> let Just x = unBox b in x
    
    1079
    +
    
    1080
    +the two `unBox b`s carry different SourceNote spans. Without tick stripping
    
    1081
    +the CoreMap treats them as distinct expressions. Long-distance information
    
    1082
    +from the outer `Just _` branch therefore never reaches the let-pattern, and
    
    1083
    +`Just x = unBox b` is wrongly reported as non-exhaustive.
    
    1064 1084
     -}
    
    1065 1085
     
    
    1066 1086
     {- Note [The Pos/Neg invariant]
    

  • testsuite/tests/pmcheck/should_compile/T27314.hs
    1
    +module T27314 where
    
    2
    +
    
    3
    +data Box = Box { unBox :: Maybe Int }
    
    4
    +
    
    5
    +f :: Box -> Int
    
    6
    +f b = case unBox b of
    
    7
    +  Nothing -> 0
    
    8
    +  Just _  -> let Just x = unBox b in x

  • testsuite/tests/pmcheck/should_compile/all.T
    ... ... @@ -93,6 +93,14 @@ test('T21360', normal, compile, [overlapping_incomplete+'-Wincomplete-record-upd
    93 93
     test('T21360b', normal, compile, [overlapping_incomplete+'-Wincomplete-record-updates'])
    
    94 94
     test('T23520', normal, compile, [overlapping_incomplete+'-Wincomplete-record-updates'])
    
    95 95
     test('T25164', [extra_files(['T25164_aux.hs']), req_th], multimod_compile, ['T25164', '-v0'])
    
    96
    +test(
    
    97
    +    'T27314',
    
    98
    +    [ omit_ways(llvm_ways), # -finfo-table-map does not work with -fllvm (#26435)
    
    99
    +      when(js_arch(), skip) # javascript doesn't support -finfo-table-map yet and yields a warning we don't want to handle here
    
    100
    +    ],
    
    101
    +    compile,
    
    102
    +    ['-Wincomplete-uni-patterns -finfo-table-map']
    
    103
    +)
    
    96 104
     
    
    97 105
     # Other tests
    
    98 106
     test('pmc001', [], compile, [overlapping_incomplete])