[GHC] #14521: Infinite loop at runtime when either : a given function is not marked INLINE, or functions are stored in strict field

[GHC]

#14521: Infinite loop at runtime when either : a given function is not marked
INLINE, or functions are stored in strict field
-------------------------------------+-------------------------------------
        Reporter:  OlivierSohn       |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.2.3
       Component:  Compiler          |              Version:  8.0.2
      Resolution:                    |             Keywords:
Operating System:  MacOS X           |         Architecture:
 Type of failure:  Incorrect result  |  Unknown/Multiple
  at runtime                         |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------
Changes (by bgamari):

 * milestone:   => 8.2.3


Old description:

> Hello,
>
> in https://github.com/OlivierSohn/hamazed/issues/1 I describe the
> following issue:
>
> When compiling on OSX with optimizations (`stack clean && stack build`
> using resolver `lts-9.12` (ghc 8.0.2)), the program loops infinitely
> (400% CPU, and execution is blocked) when an animation is triggered in
> the game. When compiling without optimizations, there is not this bug.
>
> The bug is visible at this commit :
> https://github.com/OlivierSohn/hamazed/commit/9f25223ef0502f91cd9633654bdb172f714c3920
> (to reproduce, shoot at a number in the game)
>
> I originally fixed this behaviour by pragma-declaring INLINE the function
> `Animation.animate' (this function consumes an Animator record) :
> https://github.com/OlivierSohn/hamazed/commit/597619bb14974d2bbacfb284a9e276a7cf2d2f52
>
> And later, I found that another way to fix the behaviour is to remove the
> strict annotation on the fields of the record "Animator", ie changing
>
> `
> data Animator a = Animator {
>     _animatorPure :: !(Iteration -> (Coords -> Location) -> Tree -> Tree)
>   , _animatorIO   :: !(Tree -> StepType -> Animation -> (Coords ->
> Location) -> RenderState -> IO (Maybe Animation))
>   , _animatorColorFromFrame :: !(Frame -> Color8Code)
> }
> `
>
> to:
>
> `
> data Animator a = Animator {
>     _animatorPure :: (Iteration -> (Coords -> Location) -> Tree -> Tree)
>   , _animatorIO   :: (Tree -> StepType -> Animation -> (Coords ->
> Location) -> RenderState -> IO (Maybe Animation))
>   , _animatorColorFromFrame :: (Frame -> Color8Code)
> }
>
> Could this be a compiler bug?
>
> The code is available at https://github.com/OlivierSohn/hamazed
>
> I could try to create another program that reproduces the issue more
> easily (without having to play the game), just let me know if you need
> it.
>
> Also, here is my stack version if it matters:
> `
> stack version: `1.3.2, Git revision
> 3f675146590da4f3edf768b89355f798229da2a5 (4395 commits) x86_64
> hpack-0.15.0`
>

> Thank you,
> Olivier

New description:

 Hello,

 in https://github.com/OlivierSohn/hamazed/issues/1 I describe the
 following issue:

 When compiling on OSX with optimizations (`stack clean && stack build`
 using resolver `lts-9.12` (ghc 8.0.2)), the program loops infinitely (400%
 CPU, and execution is blocked) when an animation is triggered in the game.
 When compiling without optimizations, there is not this bug.

 The bug is visible at this commit :
 https://github.com/OlivierSohn/hamazed/commit/9f25223ef0502f91cd9633654bdb172f714c3920
 (to reproduce, shoot at a number in the game)

 I originally fixed this behaviour by pragma-declaring INLINE the function
 `Animation.animate' (this function consumes an Animator record) :
 https://github.com/OlivierSohn/hamazed/commit/597619bb14974d2bbacfb284a9e276a7cf2d2f52

 And later, I found that another way to fix the behaviour is to remove the
 strict annotation on the fields of the record "Animator", ie changing

 {{{
 data Animator a = Animator {
     _animatorPure :: !(Iteration -> (Coords -> Location) -> Tree -> Tree)
   , _animatorIO   :: !(Tree -> StepType -> Animation -> (Coords ->
 Location) -> RenderState -> IO (Maybe Animation))
   , _animatorColorFromFrame :: !(Frame -> Color8Code)
 }
 }}}

 to:

 {{{
 data Animator a = Animator {
     _animatorPure :: (Iteration -> (Coords -> Location) -> Tree -> Tree)
   , _animatorIO   :: (Tree -> StepType -> Animation -> (Coords ->
 Location) -> RenderState -> IO (Maybe Animation))
   , _animatorColorFromFrame :: (Frame -> Color8Code)
 }
 }}}

 Could this be a compiler bug?

 The code is available at https://github.com/OlivierSohn/hamazed

 I could try to create another program that reproduces the issue more
 easily (without having to play the game), just let me know if you need it.

 Also, here is my stack version if it matters:
 {{{
 stack version: `1.3.2, Git revision
 3f675146590da4f3edf768b89355f798229da2a5 (4395 commits) x86_64
 hpack-0.15.0`
 }}}

 Thank you,
 Olivier

--

Comment:

 For what it's worth the repo build for me with 8.2.1,
 {{{
 $ git clone ​https://github.com/OlivierSohn/hamazed
 $ cd hamazed
 $ git checkout 9f25223ef0502f91cd9633654bdb172f714c3920
 $ git clone https://github.com/OlivierSohn/ansi-terminal.git
 $ git -C ansi-terminal checkout e6a2b1ff2e1aebd902c1791583b80ef481f370c5
 $ echo "packages: ., ansi-terminal" >> cabal.project
 $ cabal new-build all --allow-newer
 }}}
 Unfortunately I'm quite lost regarding the game itself. I found that
 pressing "Enter" makes numbers start flying about; however the `s`, `e`,
 `d`, `f` keys don't appear to do anything. This appears to be the case
 with both 8.2.2 and 8.0.2.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14521#comment:5>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler