October 2014 - ghc-tickets

[GHC] #9161: Pattern synonyms interact badly with data kinds
by GHC 17 Dec '14

17 Dec '14

#9161: Pattern synonyms interact badly with data kinds -------------------------------------------------+------------------------- Reporter: Iceland_jack | Owner: Type: bug | Status: new Priority: lowest | Milestone: Component: Compiler | Version: Keywords: renamer, pattern synonyms, | 7.8.2 data kinds | Operating System: Architecture: Unknown/Multiple | Linux Difficulty: Unknown | Type of failure: Blocked By: | None/Unknown Related Tickets: | Test Case: | Blocking: -------------------------------------------------+------------------------- If we define a pattern `A` and use it as a type: {{{ {-# LANGUAGE PatternSynonyms, DataKinds #-} pattern A = () b :: A b = undefined }}} we get the following error: {{{ Prelude> :load tmp.XDGqzccOAV.hs [1 of 1] Compiling Main ( /tmp/tmp.XDGqzccOAV.hs, interpreted ) /tmp/tmp.XDGqzccOAV.hs:5:9: GHC internal error: ‘A’ is not in scope during type checking, but it passed the renamer tcl_env of environment: [] In the type signature for ‘blah’: blah :: A Failed, modules loaded: none. }}} Also occurs with different arities of `A` (the arity of the pattern and type can differ) and defining it as an operator (enabling the `TypeOperators` pragma). -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9161> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 10

[GHC] #9226: Internal error when using equality constraint in pattern synonyms
by GHC 17 Dec '14

17 Dec '14

#9226: Internal error when using equality constraint in pattern synonyms -------------------------------------+------------------------------------- Reporter: Iceland_jack | Owner: Type: bug | Status: new Priority: lowest | Milestone: Component: Compiler | Version: 7.8.2 Keywords: renamer, pattern | Operating System: Linux synonyms, GADTs | Type of failure: GHC rejects Architecture: x86 | valid program Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | -------------------------------------+------------------------------------- While going through [http://www.andres-loeh.de/dgp2005.pdf Extensible datatypes] I had the following code: {{{ {-# LANGUAGE GADTs, PatternSynonyms, ScopedTypeVariables, Rank2Types #-} data Rep a where RUnit :: Rep () RInt :: Rep Int RChar :: Rep Char REither :: Rep a -> Rep b -> Rep (Either a b) RPair :: Rep a -> Rep b -> Rep (a, b) foo :: (Rep a, a) -> Bool foo (RUnit, ()) = True }}} And I wanted to make a pattern synonyms matching `(RUnit, ())` in `foo`, my initial attempt was: {{{ pattern PUnit1 = (RUnit, ()) }}} gives the error (`punit1.log`) while writing {{{ pattern PUnit2 = (RUnit, ()) :: (Rep (), ()) }}} compiles but can't be used in `foo`. Trying: {{{ pattern PUnit3 = (RUnit, ()) :: (a ~ ()) => (Rep a, a) }}} gives a ''GHC internal error'' (`punit3.log`). I tried various other patterns such as: {{{ pattern PUnit4 <- (RUnit, ()) pattern PUnit5 <- (RUnit, ()) :: (Rep (), ()) pattern PUnit6 <- (RUnit, ()) :: (a ~ ()) => (Rep a, a) pattern PUnit7 = (RUnit, ()) :: forall a. (a ~ ()) => (Rep a, a) pattern PUnit8 <- (RUnit, ()) :: forall a. (a ~ ()) => (Rep a, a) ... }}} None of which work. The final goal was to rewrite: {{{ eq :: Rep a -> a -> a -> Bool eq rep a b = case (rep, a, b) of (RUnit, (), ()) -> True (RInt, n1, n2) -> n1 == n2 (RChar, c1, c2) -> c1 == c2 (REither l r, Left a, Left b) -> eq l a b (REither l r, Right a, Right b) -> eq r a b (REither{}, _, _) -> False (RPair l r, (a1, a2), (b1, b2)) -> eq l a1 b1 && eq r a2 b2 }}} as something like {{{ pattern PUnit = (RUnit, ()) -- (a ~ ()) => (Rep a, a) pattern PInt n = (RInt, n) -- (a ~ Int) => (Rep a, a) -- ... eq :: Rep a -> a -> a -> Bool eq rep a b = case ((rep, a), (rep, b)) of (PUnit, PUnit) -> True (PInt n1, PInt n2) -> n1 == n2 (PChar c1, PChar c2) -> c1 == c2 (PLeft l l1, PLeft l l2) -> eq l l1 l2 (PRight r r1, PRight r r2) -> eq r r1 r2 ((REither{}, _), _) -> False (PFst l (a1, b1), PSnd r (a2, b2)) -> eq l a1 b1 && eq r a2 b2 }}} where one can avoid pattern matching against the `Rep` constructor directly. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9226> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 6

[GHC] #9353: prefetch primops are not currently useful
by GHC 15 Dec '14

15 Dec '14

#9353: prefetch primops are not currently useful -------------------------------------+------------------------------------- Reporter: MikeIzbicki | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.8.2 Keywords: | Operating System: Architecture: Unknown/Multiple | Unknown/Multiple Difficulty: Unknown | Type of failure: Blocked By: | None/Unknown Related Tickets: | Test Case: | Blocking: | Differential Revisions: -------------------------------------+------------------------------------- I wanted to use GHC's new prefetch primops to speed up a cover tree data structure I've been working on. The current implementation, however, is insufficient. The pure primops do not work at all. I propose an alternative using a seq-like syntax. ------- **The problem** Consider the function: {{{ prefetchAddr3# :: Addr# -> Int# -> Addr# }}} It is used like: {{{ -- addr# :: Addr# let prefetchAddr# = prefetchAddr3# addr# 0 ... do some stuff not involving addr# doRealWork prefetchAddr# }}} The problem is that we want the prefetch operation to get inserted at the let statement before "do some stuff...", but it doesn't get inserted there. It gets inserted directly before the call to doRealWork. This doesn't give the prefetch enough time to actually put the memory in cache, and so we still get a cache miss. I made several attempts to get around this, but they all failed due to dead code elimination. The `prefetchMutableByteArray#` functions solve this problem by incorporating an explicit state parameter. This forces the compiler to insert the prefetch operation at the location it is actually called in the code. This function, however, can only be used within the ST and IO monads, so it is of limited use. ------ ** The solution** The solution is to restructure the pure primops to use a seq-like format. For example, the prefetchAddr3# function above would become: {{{ prefetchAddr3# :: Addr# -> a -> a }}} Then we would call the function like: {{{ prefetchAddr3# addr# someOtherOp ... do some stuff not involving addr# doRealWork prefetchAddr# }}} This would correctly insert the prefetch instruction at the location it appears in the haskell code. In particular, the prefetch will occur whenever the second parameter is evaluated. This format has the advantage that it can work in non-monadic code. In my cover tree structure, I've implemented searching the tree as a `fold` operation. Every time the fold branches, only one of the branches is guaranteed to be in the cache. So I get a cache miss when I go down the other branches. I want to use the prefetch primop to prefetch the next branch the fold will go down. The fold function is pure, and I don't want to have to wedge it into the ST monad just for prefetching. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9353> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 14

[GHC] #9132: takeWhile&C. still not fusible
by GHC 15 Dec '14

15 Dec '14

#9132: takeWhile&C. still not fusible -------------------------------------+------------------------------------- Reporter: Blaisorblade | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: libraries/base | Version: 7.8.2 Keywords: fusion | Operating System: Unknown/Multiple Architecture: Unknown/Multiple | Type of failure: Runtime Difficulty: Moderate (less | performance bug than a day) | Test Case: Blocked By: | Blocking: Related Tickets: | -------------------------------------+------------------------------------- takeWhile is not still fusible, 2 1/2 years after this report: http://www.haskell.org/pipermail/glasgow-haskell- users/2011-December/021299.html The discussion suggests making takeWhile a good producer/consumer with foldr/build fusion: takeWhile' :: (a -> Bool) -> [a] -> [a] takeWhile' p xs = build $ \c n -> foldr (takeWhileF p c n) n xs {-# INLINE takeWhile' #-} takeWhileF p c n x xs = if p x then x `c` xs else n Furthermore, the discussion suggests having rewrite rules to go to this version and then rewrite back (if fusion fails). The report also mentions concatMap (which is a separate known problem). It also mentions drop and dropWhile, but I don't see when they perform allocations, so I think that's an erroneous request. I experienced the bug on GHC 7.6.3, but it seems still there in base-4.7.0.0, judging from the source: http://hackage.haskell.org/package/base-4.7.0.0/docs/src/GHC- List.html#takeWhile -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9132> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 26

[GHC] #9169: Add mkWeakTMVar to Control.Concurrent.STM.TMVar
by GHC 15 Dec '14

15 Dec '14

#9169: Add mkWeakTMVar to Control.Concurrent.STM.TMVar -------------------------------------+------------------------------------- Reporter: basvandijk | Owner: Type: feature request | Status: new Priority: normal | Milestone: Component: libraries (other) | Version: 7.8.2 Keywords: stm | Operating System: Unknown/Multiple Architecture: Unknown/Multiple | Type of failure: None/Unknown Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | -------------------------------------+------------------------------------- I just needed a Weak pointer to a TMVar: {{{ -- | Make a 'Weak' pointer to a 'TMVar', using the second argument as -- a finalizer to run when 'TMVar' is garbage-collected. mkWeakTMVar :: TMVar a -> IO () -> IO (Weak (TMVar a)) mkWeakTMVar tmv@(TMVar (TVar t#)) f = IO $ \s -> case mkWeak# t# tmv f s of (# s1, w #) -> (# s1, Weak w #) }}} It might make sense to add a similar function for TSem as well. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9169> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 7

[GHC] #9393: execvpe should handle ENOTDIR
by GHC 15 Dec '14

15 Dec '14

#9393: execvpe should handle ENOTDIR -------------------------------------+------------------------------------- Reporter: iquiw | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: libraries/unix | Version: 7.8.2 Keywords: | Operating System: Architecture: Unknown/Multiple | Unknown/Multiple Difficulty: Unknown | Type of failure: Blocked By: | None/Unknown Related Tickets: | Test Case: | Blocking: | Differential Revisions: -------------------------------------+------------------------------------- If PATH environment variable contains non directory component, execvpe fails by ENOTDIR. For example, if "/tmp/foo" is a regular file, the following program fails with PATH contains "/tmp/foo". {{{ module Main where import System.Environment import System.Posix.Process main :: IO () main = do env <- getEnvironment executeFile "echo" True [] (Just env) return () }}} {{{ $ PATH=/tmp/foo:$PATH runghc a.hs a.hs: echo: executeFile: inappropriate type (Not a directory) }}} See for example, https://github.com/haskell/cabal/issues/1723 -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9393> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 5

[GHC] #8778: Typeable TypeNats
by GHC 15 Dec '14

15 Dec '14

#8778: Typeable TypeNats -------------------------------------------+------------------------------- Reporter: dmcclean | Owner: Type: feature request | Status: new Priority: normal | Milestone: ⊥ Component: Compiler (Type checker) | Version: 7.8.1-rc1 Keywords: | Operating System: Architecture: Unknown/Multiple | Unknown/Multiple Difficulty: Unknown | Type of failure: Blocked By: | None/Unknown Related Tickets: 4385 | Test Case: | Blocking: -------------------------------------------+------------------------------- It would be useful (the case I have at hand is for some scenarios involving checking of physical dimensions) to be able to combine the Data.Dynamic story with the GHC.TypeLits story. A Typeable instance for every Nat is the sticking point. (I do not know if this is even theoretically possible.) -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8778> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 28

[GHC] #9371: Overlapping type families, segafult
by GHC 15 Dec '14

15 Dec '14

#9371: Overlapping type families, segafult ----------------------------------+---------------------------------------- Reporter: pingu | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.8.3 Keywords: | Operating System: Linux Architecture: x86_64 (amd64) | Type of failure: Runtime crash Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Revisions: ----------------------------------+---------------------------------------- Not entirely sure what's going on here. I don't think this should type check; it appears to segfault whilst calling show on the wrong type. This is probably not the absolute minimum required to reproduce. I have reproduced on 7.8.3 and 7.9.20140727. {{{#!haskell {-# LANGUAGE TypeFamilies #-} {-# LANGUAGE FlexibleInstances #-} {-# LANGUAGE UndecidableInstances #-} {-# LANGUAGE OverlappingInstances #-} import Data.Monoid class C x where data D x :: * makeD :: D x instance Monoid x => C x where data D x = D1 (Either x ()) makeD = D1 (Left mempty) instance (Monoid x, Monoid y) => C (x, y) where data D (x,y) = D2 (x,y) makeD = D2 (mempty, mempty) instance Show x => Show (D x) where show (D1 x) = show x main = print (makeD :: D (String, String)) }}} This does not segfault if you add: {{{#!haskell instance (Show x, Show y) => Show (D (x,y)) where show (D2 x) = show x }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9371> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 12

[GHC] #9415: Superclass cycle with ambiguous type causes loop
by GHC 15 Dec '14

15 Dec '14

#9415: Superclass cycle with ambiguous type causes loop -------------------------------------+------------------------------------- Reporter: goldfire | Owner: goldfire Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.9 Keywords: | Operating System: Architecture: Unknown/Multiple | Unknown/Multiple Difficulty: Unknown | Type of failure: Blocked By: | None/Unknown Related Tickets: | Test Case: | Blocking: | Differential Revisions: -------------------------------------+------------------------------------- If I say {{{ class D a => C a where meth :: D a => () class C a => D a }}} I get a loop in the typechecker. I know what's going on here: the error for the superclass cycle is added during the validity check, but then we go on to do ambiguity checks. Unfortunately, the ambiguity check never finishes. We just need to bail when there are superclass errors before doing the ambiguity check. Patch on the way. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9415> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 4

[GHC] #9390: Inlining prevents evaluation of ignored parts of unboxed tuples
by GHC 15 Dec '14

15 Dec '14

#9390: Inlining prevents evaluation of ignored parts of unboxed tuples -------------------------------------+------------------------------------- Reporter: snoyberg | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.8.3 Keywords: | Operating System: Linux Architecture: x86_64 (amd64) | Type of failure: Incorrect Difficulty: Unknown | result at runtime Blocked By: | Test Case: Related Tickets: | Blocking: | Differential Revisions: -------------------------------------+------------------------------------- Consider the following code: {{{#!hs {-# LANGUAGE MagicHash, UnboxedTuples #-} import GHC.IO (IO (..)) import GHC.Prim writeB :: MutableArray# RealWorld Char -> IO () writeB arr# = IO $ \s0# -> (# writeArray# arr# 0# 'B' s0#, () #) inlineWriteB :: MutableArray# RealWorld Char -> () inlineWriteB arr# = case f realWorld# of (# _, x #) -> x where IO f = writeB arr# test :: IO Char test = IO $ \s0# -> case newArray# 1# 'A' s0# of (# s1#, arr# #) -> case seq# (inlineWriteB arr#) s1# of (# s2#, () #) -> readArray# arr# 0# s2# main :: IO () main = test >>= print }}} I would expect this code to output the letter 'B'. When compiled without optimizations, that's exactly what it does. However, with optimizations turned on, it seems that it decides that, in `inlineWriteB`, the state value does not need to be evaluated, which results in the `writeArray#` call never occurring. This affected me when working with the vector and primitive packages. I believe I have a workaround in place (see https://github.com/haskell/primitive/pull/11) but this should probably be fixed in GHC as well. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9390> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 37