#14285: Entered absent arg - triggered by INLINEABLE, regression from 8.0.2 ---------------------------------+-------------------------------------- Reporter: j.waldmann | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.2.1 Resolution: | Keywords: Operating System: Linux | Architecture: x86_64 (amd64) Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | ---------------------------------+-------------------------------------- Comment (by j.waldmann): No. - After a clean checkout, I did {{{ stack build --dependencies-only stack exec -- ghc -isrc -fforce-recomp -O1 -dcore-lint pure-matchbox.hs ./pure-matchbox -i -w Count data/z001.srs pure-matchbox: Oops! Entered absent arg ww Map k (Set k1) }}} If I leave out the "-O1", or if I put "-O0", the error does not happen. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14285#comment:2> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

#14285: Entered absent arg - triggered by INLINEABLE, regression from 8.0.2 ---------------------------------+-------------------------------------- Reporter: j.waldmann | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.2.1 Resolution: | Keywords: Operating System: Linux | Architecture: x86_64 (amd64) Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | ---------------------------------+-------------------------------------- Comment (by RyanGlScott): By sheer dumb luck, I managed to reduce this down to the following two files: {{{#!hs module Foo where import qualified Data.Foldable as F import qualified Data.IntMap as IM import qualified Data.IntSet as IS import Prelude hiding (null) import System.Environment data Set k = Set IS.IntSet null (Set a) = IS.null a empty = Set IS.empty sfromList :: (Enum a, Foldable c) => c a -> Set a sfromList xs = Set $ IS.fromList $ Prelude.map fromEnum $ F.toList xs newtype Map k v = Map { unMap :: (IM.IntMap v) } deriving (Eq, Ord) {-# inlineable fromList #-} fromList :: Enum k => [(k,v)] -> Map k v fromList kvs = Map $ IM.fromList $ Prelude.map (\(k,v) -> (fromEnum k, v)) kvs {-# inlineable findWithDefault #-} findWithDefault d k (Map m) = IM.findWithDefault d (fromEnum k) m data Rel a b = Rel !(Map a (Set b)) !(Map b (Set a)) {-# INLINEABLE images #-} images x (Rel f b) = findWithDefault empty x f {-# INLINEABLE pre_images #-} pre_images x rel = images x $ mirrorRel rel {-# INLINEABLE mirrorRel #-} mirrorRel :: Rel a b -> Rel b a mirrorRel (Rel f g) = Rel g f }}} {{{#!hs module Main where import Foo import Prelude hiding (null) main :: IO () main = do let args = "hw" print $ null $ pre_images 'a' (Rel (fromList [('a',sfromList args)]) (fromList [('b',sfromList ar gs)])) }}} This works on GHC 8.0.2: {{{ $ /opt/ghc/8.0.2/bin/ghc -O2 -fforce-recomp Main.hs [1 of 2] Compiling Foo ( Foo.hs, Foo.o ) [2 of 2] Compiling Main ( Main.hs, Main.o ) Linking Main ... $ ./Main True }}} But not on GHC 8.2.1: {{{ $ /opt/ghc/8.0.2/bin/ghc -O2 -fforce-recomp Main.hs [1 of 2] Compiling Foo ( Foo.hs, Foo.o ) [2 of 2] Compiling Main ( Main.hs, Main.o ) Linking Main ... $ ./Main True }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14285#comment:4> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

#14285: Entered absent arg - triggered by INLINEABLE, regression from 8.0.2 ---------------------------------+-------------------------------------- Reporter: j.waldmann | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.2.1 Resolution: | Keywords: Operating System: Linux | Architecture: x86_64 (amd64) Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | ---------------------------------+-------------------------------------- Comment (by simonpj): I know what is going on. It's extremely annoying. In module `Foo` we get this after demand analysis: {{{ pre_images [InlPrag=INLINABLE] :: forall a k k. Enum a => a -> Rel k k -> Set k [Str=<S(LLLC(S(S))LLLL),1*U(A,A,A,1*C1(U(U)),A,A,A,A)> <L,U> <S(LS),1*U(A,U)>, <----------- NB Unf=Unf{Src=InlineStable, TopLvl=True, Value=True, ConLike=True, WorkFree=True, Expandable=True, Guidance=IF_ARGS [30 0 20] 120 0 Tmpl= \ (@ a_a52c) (@ k_a52e) (@ k_a52d) ($dEnum_a52n [Occ=Once] :: Enum a_a52c) (x_a2k0 [Occ=Once] :: a_a52c) (rel_a2k1 [Occ=Once!] :: Rel k_a52e k_a52d) -> case rel_a2k1 of { Rel f_a2k2 [Occ=Once] g_a2k3 [Occ=Once] -> NB -------> case T14285a.$WRel @ k_a52d @ k_a52e g_a2k3 f_a2k2 of { Rel f_a2jY [Occ=Once] _ [Occ=Dead] -> IM.findWithDefault @ (Set k_a52e) (empty @ k_a52e) (fromEnum @ a_a52c $dEnum_a52n x_a2k0) (f_a2jY `cast` (T14285a.N:Map[0] <k_a52d>_P <Set k_a52e>_N :: (Map k_a52d (Set k_a52e) :: *) ~R# (IM.IntMap (Set k_a52e) :: *))) } }}] pre_images = \ (@ a_a52c) (@ k_a52e) (@ k_a52d) ($dEnum_a52n [Dmd=<S(LLLC(S(S))LLLL),1*U(A,A,A,1*C1(U(U)),A,A,A,A)>] :: Enum a_a52c) (x_a2k0 :: a_a52c) (rel_a2k1 [Dmd=<S(LS),1*U(A,U)>] :: Rel k_a52e k_a52d) -> case rel_a2k1 of { Rel f_a2k2 [Dmd=<L,A>] g_a2k3 [Dmd=<S,1*U>] -> case fromEnum @ a_a52c $dEnum_a52n x_a2k0 of { GHC.Types.I# ww1_a57r [Dmd=<S,U>] -> Data.IntMap.Internal.$wfindWithDefault @ (Set k_a52e) (empty @ k_a52e) ww1_a57r (g_a2k3 `cast` (T14285a.N:Map[0] <k_a52d>_P <Set k_a52e>_N :: (Map k_a52d (Set k_a52e) :: *) ~R# (IM.IntMap (Set k_a52e) :: *))) } } }}} Note that * The third arg is marked `<S(LS),1*U(A,U)>`, and so is strict, but its first component is unused. * And indeed `f_a2k2` is unsed in the body of `pre_images` * But alas, in the stable-unfolding, `f_a2k2` '''is''' used. It is passed to `$WRel`, the wrapper for the strict data contructor `Rel`; it evaluates both arguments. * So if we w/w this function, we won't pass the first component; instead we'll make up `absentError "blah"` to fill the hole, expecting it not to be used. * Alas, when we do the same thing to the stable unfolding (see `Note [Worker-wrapper for INLINABLE functions]` in `WorkWrap.hs`) we ''do'' evaluate that `absentError` call. Sigh. I'm not at all clear what to do about this, but at least we can see what is going on. It's very much a corner case, so I don't want to harm mainstream cases for the sake of this one. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14285#comment:6> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

#14285: Entered absent arg - triggered by INLINEABLE, regression from 8.0.2 -------------------------------------+------------------------------------- Reporter: j.waldmann | Owner: (none) Type: bug | Status: merge Priority: normal | Milestone: 8.2.2 Component: Compiler | Version: 8.2.1 Resolution: | Keywords: Operating System: Linux | Architecture: x86_64 | (amd64) Type of failure: None/Unknown | Test Case: | stranal/should_run/T14285 Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Changes (by simonpj): * status: new => merge * testcase: => stranal/should_run/T14285 * milestone: => 8.2.2 Comment: I can't say I'm proud of this fix, but it certainly fixes it. Which is important. Could merge to future 8.2 -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14285#comment:8> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler