November 2017 - ghc-tickets

[GHC] #12159: Record-like GADTs with repeated fields (of same type) rejected
by GHC 22 Nov '17

22 Nov '17

#12159: Record-like GADTs with repeated fields (of same type) rejected -------------------------------------+------------------------------------- Reporter: heisenbug | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: GHC rejects Unknown/Multiple | valid program Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- I came across a curious bug with record-like GADTs and repeated fields. Consider following code: {{{#!hs {-# LANGUAGE GADTs #-} data Foo p where Bar :: { quux' :: Bool } -> Foo Char Baz :: { quux'' :: Bool } -> Foo Int quux :: Foo p -> Bool quux (Bar q) = q quux (Baz q) = q quuxSetter :: Foo p -> Bool -> Foo p quuxSetter old@Bar{} q = old{quux' = q} quuxSetter old@Baz{} q = old{quux'' = q} }}} This compiles and all is fine. *But* GHC is supposed to create the nice `quux` and `quuxSetter` accessors for me, right? So, let's try: {{{#!hs data Foo p where Bar :: { quux :: Bool } -> Foo Char Baz :: { quux :: Bool } -> Foo Int }}} It does not compile! Instead I get: {{{ T12159.hs:3:1: error: Constructors Bar and Baz have a common field quux , but have different result types In the data type declaration for Foo Failed, modules loaded: none. }}} This is not very polite :-) It should simply create the accessors like I did above. It obviously can be done! Testcase is attached. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/12159> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 11

[GHC] #14439: Remove redundant subtraction in (^) and stimes
by GHC 21 Nov '17

21 Nov '17

#14439: Remove redundant subtraction in (^) and stimes -------------------------------------+------------------------------------- Reporter: Bodigrim | Owner: (none) Type: task | Status: new Priority: normal | Milestone: Component: | Version: 8.2.1 libraries/base | Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- Here is the source code of `Data.Real.^`: {{{#!hs (^) :: (Num a, Integral b) => a -> b -> a x0 ^ y0 | y0 < 0 = errorWithoutStackTrace "Negative exponent" | y0 == 0 = 1 | otherwise = f x0 y0 where -- f : x0 ^ y0 = x ^ y f x y | even y = f (x * x) (y `quot` 2) | y == 1 = x | otherwise = g (x * x) ((y - 1) `quot` 2) x -- g : x0 ^ y0 = (x ^ y) * z g x y z | even y = g (x * x) (y `quot` 2) z | y == 1 = x * z | otherwise = g (x * x) ((y - 1) `quot` 2) (x * z) }}} In both cases subtraction `y - 1` is redundant. The value of `y` is guaranteed to be positive and odd, so `(y - 1) `quot` 2 = y `quot` 2`. Same argument applies to `Data.Semigroup.Internal.stimes{Monoid,Default}`. For instance, {{{#!hs stimesMonoid :: (Integral b, Monoid a) => b -> a -> a stimesMonoid n x0 = case compare n 0 of LT -> errorWithoutStackTrace "stimesMonoid: negative multiplier" EQ -> mempty GT -> f x0 n where f x y | even y = f (x `mappend` x) (y `quot` 2) | y == 1 = x | otherwise = g (x `mappend` x) (pred y `quot` 2) x g x y z | even y = g (x `mappend` x) (y `quot` 2) z | y == 1 = x `mappend` z | otherwise = g (x `mappend` x) (pred y `quot` 2) (x `mappend` z) }}} Again, `pred y` is redundant; it is enough to divide `y` itself. My proposal is to replace `y - 1` and `pred y` by `y`. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14439> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 3

[GHC] #14437: Optimise remainders by powers of two
by GHC 21 Nov '17

21 Nov '17

#14437: Optimise remainders by powers of two -------------------------------------+------------------------------------- Reporter: Bodigrim | Owner: (none) Type: feature | Status: new request | Priority: normal | Milestone: Component: Compiler | Version: 8.2.1 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- Current GHC optimises quotients, but not remainders by powers of two. For instance, even `even :: Word -> Bool` requires division: {{{#!hs evenWord :: Word -> Bool evenWord = even }}} results in CMM, containing {{{#!c if (I64[R1 + 7] % 2 != 0) goto c2sK; else goto c2sQ; }}} My proposal is to optimise `MO_{S,U}_Rem` by powers of two in `CmmOpt.cmmMachOpFoldM`, similar to existing cases for `MO_{S,U}_Quot`. Something like this may do: {{{#!hs MO_U_Rem rep | Just _ <- exactLog2 n -> Just (cmmMachOpFold dflags (MO_And rep) [x, CmmLit (CmmInt (n - 1) rep)]) MO_S_Rem rep | Just p <- exactLog2 n -> Just (cmmMachOpFold dflags (MO_Sub rep) [x, cmmMachOpFold dflags (MO_Shl rep) [cmmMachOpFold dflags (MO_S_Quot rep) [x, y], CmmLit (CmmInt p rep)]]) }}} Function `even` is used by default definition of `stimes` (`Data.Semigroup.Internal.stimesDefault`). If `<>` is cheap, division may become a bottleneck. It is also used by `GHC.Real.^`. Here is a simple benchmark: {{{#!hs v :: Int v = maximum [ x ^ (6 :: Word) | x <- [1..100000000 :: Int] ] }}} It takes 4.5 seconds on my PC before the patch above (`-O2`) and only 2.2 seconds after. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14437> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 4

[GHC] #14488: Can't define a lens for a field with a constraint
by GHC 21 Nov '17

21 Nov '17

#14488: Can't define a lens for a field with a constraint -------------------------------------+------------------------------------- Reporter: int-index | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.2 (Type checker) | Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: GHC rejects Unknown/Multiple | valid program Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- {{{#!hs {-# LANGUAGE RankNTypes #-} type Lens' s a = forall f. Functor f => (a -> f a) -> s -> f s data T a = MkT { _tfield :: Eq a => a } tfield :: Eq a => Lens' (T a) a tfield f t = MkT <$> f (_tfield t) }}} This code compiles with GHC 8.2.1. On GHC 8.0.2 the following error is reported: {{{#!hs tfield.hs:8:22: error: • Couldn't match type ‘a’ with ‘Eq a => a’ ‘a’ is a rigid type variable bound by the type signature for: tfield :: forall a. Eq a => Lens' (T a) a at tfield.hs:7:11 Expected type: f (Eq a => a) Actual type: f a • In the second argument of ‘(<$>)’, namely ‘f (_tfield t)’ In the expression: MkT <$> f (_tfield t) In an equation for ‘tfield’: tfield f t = MkT <$> f (_tfield t) • Relevant bindings include t :: T a (bound at tfield.hs:8:10) f :: a -> f a (bound at tfield.hs:8:8) tfield :: Lens' (T a) a (bound at tfield.hs:8:1) }}} I could not find a relevant GHC ticket. Unless it's a known issue, I am going to create a Phab Diff with this code as a test case to avoid a regression in the future. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14488> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 6

[GHC] #14257: Heap profiling with ghc and hp2ps and strict function application ($!) gives samples out of sequence (regression)
by GHC 21 Nov '17

21 Nov '17

#14257: Heap profiling with ghc and hp2ps and strict function application ($!) gives samples out of sequence (regression) -------------------------------------+------------------------------------- Reporter: carlostome | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Runtime | Version: 8.2.1 System | Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: #14006 Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- The following program generates an invalid .hp file when compiled with ghc 8.2.1 but it does not when using ghc 8.0.2. {{{#!hs {-# LANGUAGE ScopedTypeVariables #-} module Main where eval :: forall a b. (a -> b -> b) -> b -> [a] -> b eval f b xs = load xs [] where load :: [a] -> [a] -> b load [] stk = unload b stk load (x:xs) stk = load xs (x : stk) unload :: b -> [a] -> b unload v [] = v unload v (x : stk) = unload ((f $! x) $! v) stk main :: IO () main = print (eval (||) False (True : replicate 10000000 False)) }}} If strict application ($!) is substituted for normal application ($) or removed then the .hp generated file is correct. For reproducing the error: {{{ ghc -O2 --make -prof -fprof-auto Example.hs -fforce-recomp ./Example +RTS -hc hp2ps -e8in -c Example.hp }}} It outputs: {{{ hp2ps: Example.hp, line 43, samples out of sequence }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14257> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 16

[GHC] #14509: Consider adding new stg_ap_* functions
by GHC 21 Nov '17

21 Nov '17

#14509: Consider adding new stg_ap_* functions -------------------------------------+------------------------------------- Reporter: dobenour | Owner: (none) Type: task | Status: new Priority: normal | Milestone: 8.4.1 Component: Runtime | Version: 8.2.1 System | Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: Runtime Unknown/Multiple | performance bug Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- Currently, the RTS has a fixed set of `stg_ap_*` functions. There are two problems with this: 1. All programs pay for these functions, even programs that don’t use them. Furthermore, it is not possible for new ones to be generated by GHC for certain cases. 2. The set included in the RTS may not be optimal. (1) is a lot of work, but dealing with (2) should be “just” a matter of hunting down which functions would be useful. I would start with ByteString, Text, and other highly optimized libraries, some of which (such as ByteString) are very widely used. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14509> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 1

[GHC] #14485: Typo in blog post.
by GHC 21 Nov '17

21 Nov '17

#14485: Typo in blog post. -------------------------------------+------------------------------------- Reporter: bmusin | Owner: (none) Type: bug | Status: new Priority: low | Milestone: Component: Documentation | Version: 8.2.1 Keywords: typo | Operating System: Unknown/Multiple Architecture: | Type of failure: Documentation Unknown/Multiple | bug Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- https://ghc.haskell.org/trac/ghc/blog/ghc-8.2.11-released Should be "developers", not "develpoers" -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14485> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 2

[GHC] #14499: Pattern synonym, (a::k) cannot be written (a::N) given (k ~ N)
by GHC 21 Nov '17

21 Nov '17

#14499: Pattern synonym, (a::k) cannot be written (a::N) given (k ~ N) -------------------------------------+------------------------------------- Reporter: Iceland_jack | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.2.1 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- I was trying to create a pattern synonym that matches `TypeRep (a::k)` when `k ~ N`, witnessing the equality and returning the representation. This works fine: {{{#!hs {-# Language PatternSynonyms, ViewPatterns, GADTs, ConstraintKinds, RankNTypes, KindSignatures, PolyKinds, ScopedTypeVariables, DataKinds, TypeInType, TypeOperators, TypeApplications #-} import Type.Reflection import Data.Kind foo :: forall (kind::k') (a::k). TypeRep kind -> TypeRep a -> Maybe (kind :~~: k, TypeRep a) foo krep rep = do HRefl <- eqTypeRep krep (typeRepKind rep) pure (HRefl, rep) data N = O | S N pattern Bloop :: forall (a::k). () => N~k => TypeRep (a::k) -> TypeRep (a::k) pattern Bloop rep <- (foo (typeRep @N) -> Just (HRefl, rep :: TypeRep (a::N))) where Bloop rep = rep }}} Note that I have annotated `rep :: TypeRep (a::N)` in the view pattern, which is the first argument to `Bloop`. Let's reflect that in the type {{{#!hs -- error: -- • Expected kind ‘N’, but ‘a’ has kind ‘k’ -- • In the first argument of ‘TypeRep’, namely ‘(a :: N)’ -- In the type ‘TypeRep (a :: N) -> TypeRep (a :: k)’ -- | -- 15 | pattern Bloop :: forall (a::k). () => N ~ k => TypeRep (a::N) -> TypeRep (a::k) -- | ^ pattern Bloop :: forall (a::k). () => N~k => TypeRep (a::N) -> TypeRep (a::k) pattern Bloop rep <- (foo (typeRep @N) -> Just (rep, HRefl)) where Bloop rep = rep }}} Oh no, GHC complains that `a::k` (not `a::N`) but we have local knowledge that `N ~ k`. I know the rules for kinds are weird so I don't know if this is intended. A way around this is quantifying over an existential variable that is heterogeneously equal to `a`: {{{#!hs pattern Bloop :: forall (a::k). () => forall (n::N). a~~n => TypeRep (n::N) -> TypeRep (a::k) }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14499> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 1

[GHC] #14493: Why is g++ necessary to compile GHC?
by GHC 20 Nov '17

20 Nov '17

#14493: Why is g++ necessary to compile GHC? -------------------------------------+------------------------------------- Reporter: bgamari | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.2.1 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- Simonpj noted after a system upgrade that GHC failed to build unless `g++` is installed. `libffi`'s `configure` otherwise failed with {{{ configure:13086: checking how to run the C++ preprocessor configure:13113: gcc -E conftest.cpp gcc: error trying to exec 'cc1plus': execvp: No such file or directory }}} Why is this necessary? -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14493> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 3

[GHC] #11721: GADT-syntax data constructors don't work well with TypeApplications
by GHC 20 Nov '17

20 Nov '17

#11721: GADT-syntax data constructors don't work well with TypeApplications -------------------------------------+------------------------------------- Reporter: goldfire | Owner: goldfire Type: bug | Status: new Priority: normal | Milestone: 8.2.1 Component: Compiler | Version: 8.1 Keywords: | Operating System: Unknown/Multiple TypeApplications | Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- Consider {{{ data X a where MkX :: b -> Proxy a -> X a }}} According to the rules around specified vs. generalized variables around `TypeApplications`, the type of `MkX` should be {{{ MkX :: forall {k} (b :: *) (a :: k). b -> Proxy a -> X a }}} A few things to note: * The `k` isn't available for `TypeApplications` (that's why it's in braces), because it is not user-written. * The `b` is quantified before the `a`, because `b` comes before `a` in the user-written type signature for `MkX`. Both of these bullets are currently violated. GHCi reports `MkX`'s type as {{{ MkX :: forall k (a :: k) b. b -> Proxy a -> X a }}} It turns out that this is a hard to fix. The problem is that GHC expects data constructors to have their universal variables followed by their existential variables, always. And yet that's violated in the desired type for `MkX`. Furthermore, given the way that GHC deals with GADT return types ("rejigging", in technical parlance), it's inconvenient to get the specified/generalized distinction correct. Given time constraints, I'm afraid fixing this all won't make it for 8.0. Happily, there is are easy-to-articulate rules governing GHC's current (wrong) behavior. In a GADT-syntax data constructor: * All kind and type variables are considered specified and available for visible type application. * Universal variables always come first, in precisely the order they appear in the tycon. Note that universals that are constrained by a GADT return type are missing from the datacon. * Existential variables come next. Their order is determined by a user- written `forall`; or, if there is none, by taking the left-to-right order in the datacon's type and doing a stable topological sort. (This stable topological sort step is the same as for other user-written type signatures.) Despite the existence of these rules, it would still be better not to have special rules for GADT-syntax data constructors. This ticket is intended to capture that eventual goal. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11721> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 16