June 2017 - ghc-tickets

[GHC] #13285: Bug in GHC.Stack.callStack when used with sections
by GHC 01 Nov '17

01 Nov '17

#13285: Bug in GHC.Stack.callStack when used with sections -------------------------------------+------------------------------------- Reporter: SimonHengel | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: Incorrect result Unknown/Multiple | at runtime Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- Since GHC {{{8.0.1}}} call stacks are not constructed consistently anymore. Specifically, no stack frames are added for call sites that use section syntax. This used to work with GHC {{{7.10.2}}}. I haven't tried with {{{7.10.3}}}. Call stacks are used in {{{hspec}}} and {{{HUnit}}} to attach location information to failing test cases. Moreover, it is somewhat common to use section syntax with {{{hspec}}}. This is why I describe the bug in the context of {{{hspec}}}. But first, I give minimal steps on how to reproduce without the need of any additional dependencies. = TL;DR Minimal steps to reproduce {{{ #!haskell -- Main.hs import GHC.Stack main :: IO () main = do foo 23 42 (`foo` 23) 42 foo :: HasCallStack => Int -> Int -> IO () foo _ _ = print (length . getCallStack $ callStack) }}} {{{ $ runhaskell Main.hs }}} expected output: {{{ 1 1 }}} actual output: {{{ 1 0 }}} = Description of the bug from a users perspective (in the context of Hspec) This section describes the bug in the context of {{{hspec}}}. If you already understand how this bug affects users and why this is problematic you may choose to skip this section. == A working example Looking at the following example {{{ #!haskell -- Spec.hs import Test.Hspec main :: IO () main = hspec $ do it "works for my use case" $ do 23 `shouldBe` 42 }}} call stacks work as expected: {{{ $ runhaskell Spec.hs ... Failures: Spec.hs:6: 1) works for my use case expected: 42 but got: 23 ... }}} The users sees a source locations that points him to the failing test case. == A slightly modified and broken example If we rephrase the above example using section syntax we get {{{ #!haskell -- Spec.hs import Test.Hspec main :: IO () main = hspec $ do it "works for my use case" $ do (`shouldBe` 42) 23 }}} and things suddenly go bad: {{{ $ runhaskell Spec.hs ... Failures: src/Test/Hspec/Expectations.hs:91: 1) works for my use case expected: 42 but got: 23 ... }}} The user expects to see the call site of {{{shouldBe}}}, that points him to the failing test case. But instead he gets an unhelpful location that points somewhere at the implementation of {{{hspec-expectations}}}. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13285> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 5

[GHC] #12150: Compile time performance degradation on code that uses undefined/error with CallStacks
by GHC 01 Nov '17

01 Nov '17

#12150: Compile time performance degradation on code that uses undefined/error with CallStacks -------------------------------------+------------------------------------- Reporter: thomie | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: Compile-time Unknown/Multiple | performance bug Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- GHC 8 has a lot of trouble compiling the following program: {{{#!hs module Serialize where data Result a = Success a | Error String {- 80 guards ghc-7.10.3 -O : 0.3s ghc-8.0.1 -O : 1.8s -} instance Functor Result where {-# INLINE fmap #-} fmap | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f | bool = f where bool = undefined f = undefined }}} Here are some timing results, depending on the number of `| bool = f` clauses: {{{ * ghc-8.0.1 N clauses : time (s) 10 : 0.61 20 : 0.78 40 : 1.03 80 : 1.64 160 : 2.83 320 : 5.16 640 : 10.37 1280 : 21.16 * ghc-7.10.3 N clauses : time (s) 10 : 0.33 20 : 0.29 40 : 0.34 80 : 0.30 160 : 0.32 320 : 0.35 640 : 0.48 1280 : 0.80 }}} I think this compile time difference is caused by the `CallStack` changes introduced in GHC 8.0. When I use a version of `undefined` that doesn't have a CallStack, there is no difference in compile time when using GHC 7.10 or GHC 8.0. This is my implementation of `undefined` without `CallStack`: {{{ import GHC.Exception (errorCallException) import GHC.Prim (raise#) import Prelude (Char) error :: [Char] -> a error s = raise# (errorCallException s) undefined :: a undefined = error "undefined without callstack" }}} This is the quick and dirty Python script I used to generate those timing results (ghc version is hardcoded): {{{#!py import os import tempfile import time import subprocess def src(n): return ''' module Test where data Result a = Success a | Error String instance Functor Result where {{-# INLINE fmap #-}} fmap {0} where bool = undefined f = undefined '''.format('\n | bool = f' * n) tempfile = tempfile.mktemp('.hs') print('tempfile = {0}'.format(tempfile)) print('N clauses : time (s)') for i in range(8): n = 10 * 2 ** i with open(tempfile, 'w') as f: f.write(src(n)) f.flush() t0 = time.time() subprocess.call(['ghc-8.0.1', '-v0', '-O', tempfile]) t1 = time.time() print(str(n).ljust(10) + ': %.2f' % (t1 - t0)) os.remove(tempfile) }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/12150> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 24

[GHC] #11787: Infer HasCallStack where possible
by GHC 01 Nov '17

01 Nov '17

#11787: Infer HasCallStack where possible -------------------------------------+------------------------------------- Reporter: simonpj | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.10.3 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: -------------------------------------+------------------------------------- Consider {{{ Prelude> :t error error :: GHC.Stack.Types.HasCallStack => [Char] -> a Prelude> :t (\x -> error x) (\x -> error x) :: (?callStack::GHC.Stack.Types.CallStack) => [Char] -> a }}} We would prefer to get the same type for the latter as the former. Moreover, when we execute on #11376 (comment 25), `:t error` will show the more complicated type too. My suggestion: during generalisation, when we see `?callStack::CallStack`, use the built-in type synonym `HasCallStack`. Note that this should only happen if the name used for the implicit parameter is actually `?callstack`. Eric would you like to consider this? -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11787> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 6

[GHC] #11573: CallStacks show up unexpectedly in type of local binding in GHCi
by GHC 01 Nov '17

01 Nov '17

#11573: CallStacks show up unexpectedly in type of local binding in GHCi -------------------------------------+------------------------------------- Reporter: bgamari | Owner: Type: bug | Status: new Priority: high | Milestone: 8.0.1 Component: Compiler | Version: 8.0.1-rc2 (Type checker) | 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: -------------------------------------+------------------------------------- It was noticed (in a thread on ghc-devs) that introducing a CallStack- carrying term into the RHS of a let binding in GHCi produces a type which itself carries a callstack. I could be wrong but as far as I know this is a bug: the solver should never introduce a constraint that the user didn't write; it should merely provide an empty callstack in the event that a callstack constraint can't be solved. {{{ Prelude> let myList = [1, 2, 3 :: Integer] Prelude> let myList' = myList ++ undefined Prelude> :t myList myList :: [Integer] Prelude> :t myList' myList' :: (?callStack::GHC.Stack.Types.CallStack) => [Integer] }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11573> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 10

[GHC] #11298: Implicit call stack empty in instance declarations
by GHC 01 Nov '17

01 Nov '17

#11298: Implicit call stack empty in instance declarations -------------------------------------+------------------------------------- Reporter: pikajude | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.11 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: Incorrect result Unknown/Multiple | at runtime Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- Given the following code: {{{#!hs {-# LANGUAGE ImplicitParams #-} import GHC.Stack class Foo a where foo :: a -> String main = putStrLn $ foo () }}} In GHC 7.11.20151216, the following instances result in no output: {{{#!hs instance Foo () where foo () = prettyCallStack ?loc }}} {{{#!hs fooHelper = prettyCallStack ?loc instance Foo () where foo () = fooHelper }}} Though this one does: {{{#!hs fooHelper () = prettyCallStack ?loc instance Foo () where foo = fooHelper }}} The aforementioned instances all yield output with GHC 7.10.3 (after replacing `prettyCallStack` with `showCallStack`). -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11298> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 10

[GHC] #13372: Attach CallStack to IOError, add CallStack constraints to relevant functions in base
by GHC 31 Oct '17

31 Oct '17

#13372: Attach CallStack to IOError, add CallStack constraints to relevant functions in base -------------------------------------+------------------------------------- Reporter: ezyang | Owner: (none) Type: feature | Status: new request | Priority: normal | Milestone: Component: | Version: 8.1 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: -------------------------------------+------------------------------------- In #12096 it is requested that we add CallStack to SomeException. This is a more conservative request for adding CallStack to IOError only. IOError is already an abstract data type, so it is a simple matter to augment it with space for a CallStack. The main design flex point is whether or not a `CallStack` should be associated with an IOError when it is constructed, or when it is thrown. There are pros and cons to both: 1. If it is associated at construction time, the call stack will reflect when the IOError was constructed, not when it was thrown. This could be arbitrarily far away from the actual throw site. I've been mostly using CallStack for monadic code, and so my primary concern is with where in the monadic code the exception was thrown, so this info isn't really what I want. 2. If it is associated at throw time, there are a few other problems. First, if someone uses throwIO rather than ioError, we won't actually associate a CallStack with the value. Second, if a user is rethrowing an exception, they may accidentally overwrite the old stack trace; but it was the original one that I wanted. For now, I suggest going with (1). Here is the proposal: * Add `ioeGetCallStack` and `ioeSetCallStack`, for setting and getting the `CallStack` recorded in `IOError` * `userError` and `mkIOError` gain a `HasCallStack` constraint. We have a choice for `annotateIOError`, but my suggestion is to NOT annotate it. * Let's annotate all functions that (transitively) may raise an IOError in base with `HasCallStack`. The primary benefit is now if I have a "file not found" exception or similar, I automatically get a stack trace saying exactly what code was run where the file was not found. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13372> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 6

[GHC] #12143: ApplicativeDo Fails to Desugar 'return True'
by GHC 31 Oct '17

31 Oct '17

#12143: ApplicativeDo Fails to Desugar 'return True' -------------------------------------+------------------------------------- Reporter: MichaelBurge | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 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: -------------------------------------+------------------------------------- If I bring in a spurious parameter, ApplicativeDo correctly infers Functor. {{{ Prelude> :t \m -> do { x <- m; return True; } \m -> do { x <- m; return True; } :: Functor f => f t -> f Bool Prelude> :t \m -> do { m; return True; } \m -> do { m; return True; } :: Monad m => m a -> m Bool }}} However, without the spurious parameter, ApplicativeDo assumes monad. {{{ Prelude> :t do { return True; } do { return True; } :: Monad m => m Bool }}} The first block in particular shows two examples that seem inconsistent when placed next to each other. It seems like if 'x' is unused, then 'x <- m' and 'm' should be equivalent. I would expect all of these to desugar to Functor or Applicative. And maybe 'return True' should be Monad, but 'do { return True; }' should be Applicative? -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/12143> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 8

[GHC] #13222: Update formalism for join points
by GHC 30 Oct '17

30 Oct '17

#13222: Update formalism for join points -------------------------------------+------------------------------------- Reporter: lukemaurer | Owner: lukemaurer Type: task | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.1 Keywords: JoinPoints | 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 rather flagrantly violated the entreaties in `CoreLint` about updating the formalism (namely by adding join points). Need to fix that. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13222> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 8

[GHC] #13825: Allow multiple constructor fields occupy the same word
by GHC 29 Oct '17

29 Oct '17

#13825: Allow multiple constructor fields occupy the same word -------------------------------------+------------------------------------- Reporter: michalt | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 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: -------------------------------------+------------------------------------- The main goal is to reduce the overhead of things like: {{{#!hs data Bloated = Bloated {-# UNPACK #-} !Word8 {-# UNPACK #-} !Int8 {-# UNPACK #-} !Bool }}} Assuming 64-bit architecture, currently those fields will take 8 bytes each! So for this example we'd need: 8 bytes for header + 3 * 8 bytes for fields = 32 bytes. But we should be able to pack the fields into a single word (a word is 8 bytes and each field really only needs 1 byte) for a total of 16 bytes (8 bytes header + 8 bytes for fields, with the 5 bytes being "overhead" due to heap alignment). My understanding is that we need a few things to make this happen: - Ability to refer to fields that are packed into a single word (currently everything in GHC assumes that each field occupies a single word). Simon Marlow started working on this in https://phabricator.haskell.org/D38 - Introduce primitives like `Word8#`, `Int8#`, ... (currently `WordX` and `IntX` are defined as wrappers of `Word#` and `Int#` respectively) and change `WordX`/`IntX` definitions to use those primitives. - Figure out what to do with `Bool` (should it be just `Word8#`? should we have `Bool#`?) and change its definition (using pattern synonyms for `True`/`False`) Some additional info: - Thread on ghc-devs: https://mail.haskell.org/pipermail/ghc- devs/2017-June/014304.html -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13825> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 11

[GHC] #13038: implementation of Modus ponens and Modus tollens
by GHC 28 Oct '17

28 Oct '17

#13038: implementation of Modus ponens and Modus tollens -------------------------------------+------------------------------------- Reporter: vanto | Owner: Type: feature | Status: new request | Priority: normal | Milestone: Component: | Version: 8.0.1 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: -------------------------------------+------------------------------------- Can we have in Prelude - Booleans, Modus ponens and Modus tollens implemented? Or in another module? reminder:[[BR]] P is a proposition and v(p) is the logic value of P.[[BR]] P is true, v(P) = 1[[BR]] P is false, v(P) = 0[[BR]] P -> Q is always true in propositional logic. Modus ponens:[[BR]] v(P) = 1[[BR]] v(P -> Q) = 1[[BR]] v(Q) = 1[[BR]] v(P) and v(P -> q) are premiss, v(q) is the conclusion.[[BR]] or in a more general form[[BR]] P : X is A[[BR]] P -> Q : A -> B[[BR]] Q : X is B[[BR]] Modus tollens:[[BR]] v(Q) = 0[[BR]] v(P -> Q) = 1[[BR]] v(P) = 0[[BR]] v(Q) and v(P -> Q) are premiss, v(P) is the conclusion.[[BR]] or in a more general form[[BR]] not Q : X is not B[[BR]] P -> Q : A -> B[[BR]] not P : X is not A[[BR]][[BR]] As well as using the numbers 0 and 1 for the respective values false and true, we'll use the letters F for False and T for True.[[BR]] We are always:[[BR]] v(F) = 0[[BR]] v(T) = 1[[BR]] v(P -> Q) = 1[[BR]] -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13038> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 11