February 2019 - ghc-tickets

[GHC] #16288: Core Lint error: Occurrence is GlobalId, but binding is LocalId
by GHC 04 Mar '19

04 Mar '19

#16288: Core Lint error: Occurrence is GlobalId, but binding is LocalId -------------------------------------+------------------------------------- Reporter: monoidal | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.7 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: Compile-time Unknown/Multiple | crash or panic Test Case: | Blocked By: Blocking: 15840 | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- Compiling the following three modules causes a Core Lint error in HEAD. (This does not happen in 8.6 - the Lint check was introduced later.) To reproduce: save the three files in `Repro/` directory and use `ghc- stage2 -dcore-lint -O Repro/B.hs`. The reproduction code is minimized version of code from cabal and prettyprint libraries. A.hs {{{ #!haskell module Repro.A where import Repro.C data License class Pretty a where pretty :: a -> Doc instance Pretty License where pretty _ = pretV bar :: (Pretty a) => a -> Doc bar w = foo (pretty (u w w w w)) u :: a -> a -> a -> a -> a u = u }}} B.hs {{{ #!haskell module Repro.B where import Repro.A import Repro.C bar2 :: License -> Doc bar2 = bar }}} C.hs {{{ #!haskell module Repro.C where data Doc = Empty | Beside Doc hcat :: Doc -> Doc hcat Empty = Empty hcat xs = hcat xs pretV = hcat Empty foo :: Doc -> Doc foo Empty = hcat Empty foo val = Beside val }}} The error: {{{ *** Core Lint errors : in result of Simplifier *** Repro/C.hs:9:1: warning: [in body of letrec with binders pretV_r3 :: Doc] Occurrence is GlobalId, but binding is LocalId pretV :: Doc [GblId, Unf=Unf{Src=<vanilla>, TopLvl=True, Value=False, ConLike=False, WorkFree=False, Expandable=False, Guidance=IF_ARGS [] 20 0}] *** Offending Program *** lvl_s1kN :: Doc [LclId, Unf=Unf{Src=<vanilla>, TopLvl=True, Value=False, ConLike=False, WorkFree=True, Expandable=False, Guidance=IF_ARGS [] 30 20}] lvl_s1kN = case pretV of wild_Xd { Empty -> pretV; Beside ipv_s105 -> Beside wild_Xd } $sbar_s1kL [InlPrag=NOUSERINLINE[2]] :: License -> Doc [LclId, Arity=1, Unf=Unf{Src=InlineStable, TopLvl=True, Value=True, ConLike=True, WorkFree=True, Expandable=True, Guidance=ALWAYS_IF(arity=1,unsat_ok=True,boring_ok=True) Tmpl= \ _ [Occ=Dead] -> case pretV of wild_Xd [Occ=Once*] { Empty -> let { pretV_r3 :: Doc [LclId] pretV_r3 = wild_Xd } in pretV; Beside _ [Occ=Dead] -> Beside wild_Xd }}] $sbar_s1kL = \ _ [Occ=Dead] -> lvl_s1kN $trModule_s1kE :: Addr# [LclId, Unf=Unf{Src=<vanilla>, TopLvl=True, Value=True, ConLike=True, WorkFree=True, Expandable=True, Guidance=IF_ARGS [] 20 0}] $trModule_s1kE = "main"# $trModule_s1kF :: TrName [LclId, Unf=Unf{Src=<vanilla>, TopLvl=True, Value=True, ConLike=True, WorkFree=True, Expandable=True, Guidance=IF_ARGS [] 10 20}] $trModule_s1kF = TrNameS $trModule_s1kE $trModule_s1kG :: Addr# [LclId, Unf=Unf{Src=<vanilla>, TopLvl=True, Value=True, ConLike=True, WorkFree=True, Expandable=True, Guidance=IF_ARGS [] 30 0}] $trModule_s1kG = "Repro.B"# $trModule_s1kH :: TrName [LclId, Unf=Unf{Src=<vanilla>, TopLvl=True, Value=True, ConLike=True, WorkFree=True, Expandable=True, Guidance=IF_ARGS [] 10 20}] $trModule_s1kH = TrNameS $trModule_s1kG $trModule :: Module [LclIdX, Unf=Unf{Src=<vanilla>, TopLvl=True, Value=True, ConLike=True, WorkFree=True, Expandable=True, Guidance=IF_ARGS [] 10 30}] $trModule = Module $trModule_s1kF $trModule_s1kH bar2 :: License -> Doc [LclIdX, Arity=1, Unf=Unf{Src=<vanilla>, TopLvl=True, Value=True, ConLike=True, WorkFree=True, Expandable=True, Guidance=IF_ARGS [0] 30 20}] bar2 = \ _ [Occ=Dead] -> case pretV of wild_Xd { Empty -> pretV; Beside ipv_s105 -> Beside wild_Xd } *** End of Offense *** }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/16288> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 8

[GHC] #16254: INLINABLE pragma and newtype wrappers prevent inlining
by GHC 04 Mar '19

04 Mar '19

#16254: INLINABLE pragma and newtype wrappers prevent inlining -------------------------------------+------------------------------------- Reporter: monoidal | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.6.3 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: #5327 Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- In #5327, we allowed case-of-known constructor to see through bindings that were marked as INLINABLE and used newtypes. However, this works only if the newtype does not have a wrapper. The following code, which is the same as T5237 except for the extra type parameters in Size, does not cause the optimization to fire. {{{ {-# LANGUAGE GADTs, ExplicitForAll #-} module T16254 where newtype Size a b where Size :: forall b a. Int -> Size a b {-# INLINABLE val2 #-} val2 = Size 17 f n = case val2 of Size s -> s + s > n }}} The reason is that `exprIsConApp_maybe` sees `$WSize (I# 17#)`, which is an application with a nontrivial argument. In general, this could cause duplication of work, but in this case we're dealing with a newtype wrapper, so this is safe. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/16254> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 12

[GHC] #16272: libffi copies files into rts build tree
by GHC 04 Mar '19

04 Mar '19

#16272: libffi copies files into rts build tree -------------------------------------+------------------------------------- Reporter: mpickering | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Build System | Version: 8.6.3 (Hadrian) | 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 `rts` should explicitly depend on `ffi.h` and `ffitarget.h`, then use this path to copy in these two header files itself rather than relying on the libffi build rule to copy them into the `rts/build` directory. You can see this happening in `Rules.Libffi.libffiRules` This often results in a failure where the `rts` rule will fail to produce these files after deleting build products in `_build`. {{{ shakeArgsWith 0.000s 0% Function shake 0.006s 0% Database read 1.051s 0% Database compression 0.146s 0% With database 0.031s 0% Running rules 121.010s 98% ========================= Total 122.244s 100% Error when running Shake build system: at src/Main.hs:58:30-42: * Depends on: stage1:lib:text at src/Rules/SimpleTargets.hs:31:5-15: * Depends on: integersimple/stage1/lib/package.conf.d/text-1.2.3.1.conf * Depends on: OracleQ (ContextDataKey (Context {stage = Stage1, package = Package {pkgType = Library, pkgName = "text", pkgPath = "libraries/text"}, way = v})) * Depends on: integersimple/stage1/libraries/text/setup-config * Depends on: integersimple/stage1/lib/package.conf.d/base-4.12.0.0.conf * Depends on: OracleQ (ContextDataKey (Context {stage = Stage1, package = Package {pkgType = Library, pkgName = "base", pkgPath = "libraries/base"}, way = v})) * Depends on: integersimple/stage1/libraries/base/setup-config * Depends on: integersimple/stage1/lib/package.conf.d/rts-1.0.conf * Depends on: integersimple/stage1/rts/build/libHSrts-1.0_thr_l.a * Depends on: integersimple/stage1/rts/build/c/Interpreter.thr_l_o * Depends on: integersimple/stage1/rts/build/ffi.h * Depends on: integersimple/stage1/rts/build/ffi.h integersimple/stage1/rts/build/ffitarget.h * Raised the exception: Error, &%> rule failed to produce 2 files (out of 2) integersimple/stage1/rts/build/ffi.h - MISSING integersimple/stage1/rts/build/ffitarget.h - MISSING CallStack (from HasCallStack): error, called at src/Development/Shake/Internal/Rules/Files.hs:245:13 in shake-0.17.2-586622e6ae1d899ec790b1028cae3413c00a616f124bac634769a9607581e631:Development.Shake.Internal.Rules.Files }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/16272> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 3

[GHC] #16295: Enable cached builds with Hadrian
by GHC 04 Mar '19

04 Mar '19

#16295: Enable cached builds with Hadrian -------------------------------------+------------------------------------- Reporter: snowleopard | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Build System | Version: 8.6.3 (Hadrian) | 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: -------------------------------------+------------------------------------- As Neil is working on turning Shake into Cloud Shake [1], Shake has got a new feature of "locally cached builds", making it possible to reuse build artefacts from previous builds on the same machine. This means, in particular, that a clean build could be finished very quickly if you've built GHC before from the same/similar sources. Hopefully, this could also help us speed up GHC's CI some day. To enable cached builds, it is important to find and fix all instances of untracked dependencies in Hadrian, or at least record them using the `produces` feature of Shake [2], because otherwise Shake has no idea that these files are actually needed and should therefore be stored in the cache. Here are a couple of tickets related to untracked dependencies: https://ghc.haskell.org/trac/ghc/ticket/16271 https://ghc.haskell.org/trac/ghc/ticket/16272 I am currently working this and hope to produce a patch soon. [1] See the Build Systems à la Carte paper: https://dl.acm.org/ft_gateway.cfm?id=3236774 [2] http://hackage.haskell.org/package/shake/docs/Development- Shake.html#v:produces -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/16295> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 2

[GHC] #16296: OccurAnal should not break the linter assumptions
by GHC 04 Mar '19

04 Mar '19

#16296: OccurAnal should not break the linter assumptions -------------------------------------+------------------------------------- Reporter: aspiwack | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.6.3 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: -------------------------------------+------------------------------------- As discovered in #16288 (see in particular https://ghc.haskell.org/trac/ghc/ticket/16288#comment:2) the implementation of the binder-swap transformation in the occurrence analysis sometimes produces terms which would be refused by the linter. Specifically: {{{#!hs case A.x of u { pat -> rhs } }}} Becomes {{{#!hs case A.x of u { pat -> let x = u in rhs } }}} Crucially, here, `A.x` (which is a global id) and `x` (which is a local id) ''have the same unique''. This is so that `x` shadows `A.x` in `rhs` and captures all the occurrences of `A.x` in rhs. Which are now bound by `x`. But all these occurrences of `A.x` (which are now occurrences of `x`) are still marked as global ids. This is rejected by the linter. The occurrence analysis is counting on the fact that there will be a simplifier step before the next linter fires, which will inline `x` and replace it by `u` everywhere. As #16288 has shown, this is not always the case! (specifically, in #16288, occurrence analysis happens in an unfolding, but is not followed by a simplifier pass). ---- It's easy to fix such bugs: turn off binder-swap in occurrence analysis when not followed by the simplifier. But it is very fragile. So the longer term solution would be to never produce term which break the linter. The entire reason for this `let x = u` is to avoid carrying a substitution in the occurrence analysis (and offload the work of actually doing the substitution to the simplifier). But the occurrence analysis will work through the entire `rhs` anyway. So it should be able to, at very little cost, propagate a substitution. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/16296> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 3

[GHC] #16346: Lint error on master: Occurrence is GlobalId, but binding is LocalId
by GHC 04 Mar '19

04 Mar '19

#16346: Lint error on master: Occurrence is GlobalId, but binding is LocalId -------------------------------------+------------------------------------- Reporter: simonpj | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.6.3 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: -------------------------------------+------------------------------------- I'm getting this Lint error when validating 'master', when compiling `PrelNames.hs` with the stage1 compiler. {{{ *** Core Lint errors : in result of Simplifier *** <no location info>: warning: In the expression: Name (External modu_a3Ct) (OccName dt_X6Ya dt_X6Yc) dt2_a6Yf noSrcSpan Occurrence is GlobalId, but binding is LocalId noSrcSpan :: SrcSpan [GblId, Unf=Unf{Src=<vanilla>, TopLvl=True, Value=False, ConLike=False, WorkFree=True, Expandable=False, Guidance=IF_ARGS [] 10 0}] }}} How come this got past CI? -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/16346> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 10

[GHC] #8657: -fregs-graph still has a limit on spill slots
by GHC 04 Mar '19

04 Mar '19

#8657: -fregs-graph still has a limit on spill slots -----------------------------------+--------------------------------------- Reporter: simonmar | Owner: Type: bug | Status: new Priority: normal | Milestone: 7.8.1 Component: Compiler (NCG) | Version: 7.7 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: Compile-time crash Unknown/Multiple | Test Case: Difficulty: Unknown | Blocking: Blocked By: | Related Tickets: | -----------------------------------+--------------------------------------- The limit on spill slots was removed for the linear register allocator in 7.8, but not for `-fregs-graph`. SHA-1 fails to compile with `-fregs-graph` in 7.8. Related: #7679 (`-fregs-graph` generates poor code with the new codegen). -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8657> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 18

[GHC] #15471: Polymorphism, typed splices and type inference don't mix
by GHC 04 Mar '19

04 Mar '19

#15471: Polymorphism, typed splices and type inference don't mix -------------------------------------+------------------------------------- Reporter: mpickering | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: 8.6.1 Component: Compiler | Version: 8.4.3 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: -------------------------------------+------------------------------------- Trying to quote and splice a polymorphic function doesn't work unless you have a type signature. {{{ {-# LANGUAGE TemplateHaskell #-} module A where foo1 x = x test_foo = [|| foo1 ||] }}} {{{ {-# LANGUAGE TemplateHaskell #-} module B where import A qux = $$(test_foo) }}} The type of `qux` is `Any -> Any`! Which is clearly wrong. Adding a type signature to `qux` fixes the problem. {{{ qux :: a -> a qux = $$(test_foo) }}} Either there should be a better error or this should just work. It seems that this has always been broken. Ryan tested on GHC 7.8.4 for me. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15471> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 19

[GHC] #12919: Equality not used for substitution
by GHC 04 Mar '19

04 Mar '19

#12919: Equality not used for substitution -------------------------------------+------------------------------------- Reporter: int-index | Owner: Type: bug | Status: new Priority: high | Milestone: Component: Compiler | Version: 8.0.1 (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: -------------------------------------+------------------------------------- This code {{{ {-# LANGUAGE TypeInType, TypeFamilies, GADTs, ConstraintKinds #-} import Data.Kind data N = Z data V :: N -> Type where VZ :: V Z type family VC (n :: N) :: Type where VC Z = Type type family VF (xs :: V n) (f :: VC n) :: Type where VF VZ f = f data Dict c where Dict :: c => Dict c prop :: xs ~ VZ => Dict (VF xs f ~ f) prop = Dict }}} fails with this error: {{{ Op.hs:20:8: error: • Couldn't match type ‘f’ with ‘VF 'VZ f’ arising from a use of ‘Dict’ ‘f’ is a rigid type variable bound by the type signature for: prop :: forall (xs :: V 'Z) f. xs ~ 'VZ => Dict VF xs f ~ f at Op.hs:19:9 • In the expression: Dict In an equation for ‘prop’: prop = Dict • Relevant bindings include prop :: Dict VF xs f ~ f (bound at Op.hs:20:1) }}} However, if I substitute `xs` with `VZ` (by hand) in the type of `prop`, it compiles. Can be reproduced with GHC 8.0.1 but not HEAD. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/12919> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 25

[GHC] #15549: Core Lint error with empty closed type family
by GHC 04 Mar '19

04 Mar '19

#15549: Core Lint error with empty closed type family -------------------------------------+------------------------------------- Reporter: RyanGlScott | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: 8.6.1 Component: Compiler | Version: 8.4.3 (Type checker) | Keywords: TypeFamilies, | Operating System: Unknown/Multiple TypeInType | Architecture: | Type of failure: Compile-time Unknown/Multiple | crash or panic Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- The following program gives a Core Lint error on GHC 8.2.2 and later: {{{#!hs {-# LANGUAGE EmptyCase #-} {-# LANGUAGE ScopedTypeVariables #-} {-# LANGUAGE TypeFamilies #-} {-# LANGUAGE TypeInType #-} {-# LANGUAGE UndecidableInstances #-} module Bug where import Data.Kind import Data.Void data family Sing (a :: k) data V1 :: Type -> Type data instance Sing (z :: V1 p) class Generic a where type Rep a :: Type -> Type to :: Rep a x -> a class PGeneric a where type To (z :: Rep a x) :: a class SGeneric a where sTo :: forall x (r :: Rep a x). Sing r -> Sing (To r :: a) ----- instance Generic Void where type Rep Void = V1 to x = case x of {} type family EmptyCase (a :: j) :: k where instance PGeneric Void where type To x = EmptyCase x instance SGeneric Void where sTo x = case x of }}} {{{ $ /opt/ghc/8.4.3/bin/ghc Bug.hs -dcore-lint -fforce-recomp [1 of 1] Compiling Bug ( Bug.hs, Bug.o ) *** Core Lint errors : in result of Simplifier *** Bug.hs:38:7: warning: [in body of lambda with binder x_aZ8 :: Sing r_a12q] Kind application error in coercion ‘(Sing (D:R:RepVoid[0] <x_a12p>_N) <r_a12q>_N)_R’ Function kind = forall k. k -> * Arg kinds = [(V1 x_a12p, *), (r_a12q, Rep Void x_a12p)] Fun: V1 x_a12p (r_a12q, Rep Void x_a12p) Bug.hs:38:7: warning: [in body of lambda with binder x_aZ8 :: Sing r_a12q] Kind application error in coercion ‘D:R:SingV1z0[0] <x_a12p>_N <r_a12q>_N’ Function kind = V1 x_a12p -> * Arg kinds = [(r_a12q, Rep Void x_a12p)] Fun: V1 x_a12p (r_a12q, Rep Void x_a12p) Bug.hs:38:7: warning: [in body of lambda with binder x_aZ8 :: Sing r_a12q] Kind application error in coercion ‘D:R:SingV1z0[0] <x_a12p>_N <r_a12q>_N’ Function kind = V1 x_a12p -> * Arg kinds = [(r_a12q, Rep Void x_a12p)] Fun: V1 x_a12p (r_a12q, Rep Void x_a12p) <no location info>: warning: In the type ‘R:SingV1z x_a12p r_a12q’ Kind application error in type ‘R:SingV1z x_a12p r_a12q’ Function kind = forall p -> V1 p -> * Arg kinds = [(x_a12p, *), (r_a12q, Rep Void x_a12p)] Fun: V1 x_a12p (r_a12q, Rep Void x_a12p) <no location info>: warning: In the type ‘R:SingV1z x_a12p r_a12q’ Kind application error in type ‘R:SingV1z x_a12p r_a12q’ Function kind = forall p -> V1 p -> * Arg kinds = [(x_a12p, *), (r_a12q, Rep Void x_a12p)] Fun: V1 x_a12p (r_a12q, Rep Void x_a12p) *** Offending Program *** <elided> $csTo_a12n :: forall x (r :: Rep Void x). Sing r -> Sing (To r) [LclId, Arity=1, Str=<L,U>x, Unf=Unf{Src=<vanilla>, TopLvl=True, Value=True, ConLike=True, WorkFree=True, Expandable=True, Guidance=ALWAYS_IF(arity=1,unsat_ok=True,boring_ok=True)}] $csTo_a12n = \ (@ x_a12p) (@ (r_a12q :: Rep Void x_a12p)) (x_aZ8 :: Sing r_a12q) -> case x_aZ8 `cast` ((Sing (D:R:RepVoid[0] <x_a12p>_N) <r_a12q>_N)_R ; D:R:SingV1z0[0] <x_a12p>_N <r_a12q>_N :: (Sing r_a12q :: *) ~R# (R:SingV1z x_a12p r_a12q :: *)) of nt_s13T { } }}} GHC 8.0.2 does not appear to suffer from this Core Lint error. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15549> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 10