Opened 9 months ago

Closed 9 months ago

#14379 closed bug (fixed)

Regression - GHC 8.2.1 Consumes All Memory On Build

Reported by: jm4games Owned by:
Priority: high Milestone: 8.2.2
Component: Compiler Version: 8.2.1
Keywords: Cc:
Operating System: Linux Architecture: x86_64 (amd64)
Type of failure: Compile-time crash or panic Test Case:
Blocked By: Blocking:
Related Tickets: Differential Rev(s): 8.0.2
Wiki Page:

Description (last modified by jm4games)

The following code will cause GHC to consume all memory/swap and eventually crash (a regression from 8.0.2).

Code highlighting:

module Test.Test where

import Data.Text (Text)

import Data.Monoid ((<>))
import Data.Vector as V
import TextShow (showt)

compileTest :: V.Vector (Text, V.Vector (Int, V.Vector a)) -> V.Vector (Text, V.Vector (Int, V.Vector a)) -> Either Text ()
compileTest vecA vecB = V.ifoldl' validateSym (Right ()) vecB
    validateSym :: Either Text () -> Int -> (Text, V.Vector (Int, V.Vector a)) -> Either Text ()
    validateSym res iSym (sym, freqs)
      | Just sym == (fst <$> (vecA V.!? iSym)) = V.ifoldl' validateFreq res freqs
      | otherwise = Left $
            if iSym < V.length vecA then
              "Seed data" <> " not found at index [" <> showt iSym <> "]."
              else "No " <> sym <> " at index " <> showt iSym <> "."
        validateFreq :: Either Text () -> Int -> (Int, V.Vector a) -> Either Text ()
        validateFreq res2 iFreq (freq, _)
          | freq == fst (snd (vecA V.! iSym) V.! iFreq) = res2
          | otherwise = Left $
                "Seed data " <> (fst (vecA V.! iSym)) <>
                " at frequency " <> showt (fst (snd (vecA V.! iSym) V.! iFreq)) <>
                " not found at index [" <> showt iSym <> "][" <> showt iFreq -- <> "]."

NOTE: The snippet is large (and messy) because there seems to be an exact sequence of evaluation to causing the out of memory. For example if you comment out line 26 (<> showt iFreq) it will allow the code to compile. Like wise if I comment out all of line 25 it will compile. I can't seem to figure out what exact combination of things causes the issue.

Cabal file (used with stack 1.5.1, resolver: nightly-2017-10-21).

Code highlighting:

name:                some-test
build-type:          Simple
cabal-version:       >= 1.10

  default-language: Haskell2010
  ghc-options:      -Wall -fwarn-tabs -funbox-strict-fields -fno-warn-unused-do-bind -O2
  ghc-prof-options: -fprof-auto


    base                    >= 4.9   && < 4.11,
    text                    >= 1.2,
    text-show               >= 3.4   && < 3.7,
    vector                  >= 0.10  && < 0.13


Change History (10)

comment:1 Changed 9 months ago by jm4games

Description: modified (diff)
Differential Rev(s): 8.0.2
Summary: GHC 2.8.1 Consumes All Memory On BuildRegression - GHC 2.8.1 Consumes All Memory On Build

comment:2 Changed 9 months ago by simonpj

As so often, I failed to reproduce this with HEAD, at the first fence

simonpj@cam-05-unx:/cam-01-srv/simonpj/tmp/T14379$ cabal install --with-ghc=/home/simonpj/5builds/HEAD/inplace/bin/ghc-stage2 --allow-newer
Resolving dependencies...
Configuring primitive-
Building primitive-
Failed to install primitive-
Build log ( /home/simonpj/.cabal/logs/primitive- ):
cabal: Entering directory '/tmp/cabal-tmp-46331/primitive-'
Configuring primitive-
Building primitive-
Preprocessing library primitive-
[ 1 of 12] Compiling Control.Monad.Primitive ( Control/Monad/Primitive.hs, dist/build/Control/Monad/Primitive.o )

Control/Monad/Primitive.hs:45:1: warning: [-Wdeprecations]
    Module ‘Control.Monad.Trans.List’ is deprecated:
      This transformer is invalid on most monads
45 | import Control.Monad.Trans.List     ( ListT    )
   | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ many more similar deprecation warnings...

[ 2 of 12] Compiling Data.Primitive.Internal.Compat ( Data/Primitive/Internal/Compat.hs, dist/build/Data/Primitive/Internal/Compat.o )
[ 3 of 12] Compiling Data.Primitive.Array ( Data/Primitive/Array.hs, dist/build/Data/Primitive/Array.o )

Data/Primitive/Array.hs:531:10: error:
    • No instance for (Semigroup (Array a))
        arising from the superclasses of an instance declaration
    • In the instance declaration for ‘Monoid (Array a)’
531 | instance Monoid (Array a) where
    |          ^^^^^^^^^^^^^^^^
cabal: Leaving directory '/tmp/cabal-tmp-46331/primitive-'
cabal: Error: some packages failed to install:
primitive- failed during the building phase. The exception was:
ExitFailure 1
some-test- depends on primitive- which failed to install.
vector- depends on primitive- which failed to install.

Without --allow-newer it falls over much faster. Any ideas for how to fix?

I will try again with 8.2.

comment:3 Changed 9 months ago by mpickering

Since Herbert merged the commit which made Semigroup a superclass of Monoid it became much more difficult to build packages with HEAD.

The most reliable way I found now was to use the package overlay he posted about but it is more fiddly to set up.

comment:4 Changed 9 months ago by simonpj

OK I have a fix for this. It was SpecConstr going wild.

comment:5 Changed 9 months ago by Simon Peyton Jones <simonpj@…>

In 7d7d94fb/ghc:

Fix an exponential-blowup case in SpecConstr

Trac #14379 showed a case where use of "forcing" to do
"damn the torpedos" specialisation without resource limits
(which 'vector' does a lot) led to exponential blowup.

The fix is easy.  Finding it wasn't.  See Note [Forcing
specialisation] and the one-line change in decreaseSpecCount.

comment:6 Changed 9 months ago by simonpj

This would be worth merging to 8.2.

I have not added a test because the repro case is tricky.

My change deliberately reduces the amount of specialisation that SpecConstr does; but I suppose it's possible that some other important use-case will get worse. But I think it's unlikely. Worth a check on

comment:7 Changed 9 months ago by simonpj

Status: newmerge

comment:8 Changed 9 months ago by jm4games

Thanks alot simon!

comment:9 Changed 9 months ago by jm4games

Summary: Regression - GHC 2.8.1 Consumes All Memory On BuildRegression - GHC 8.2.1 Consumes All Memory On Build

comment:10 Changed 9 months ago by bgamari

Milestone: 8.2.2
Resolution: fixed
Status: mergeclosed
Note: See TracTickets for help on using tickets.