bounds error on sparc when compiling haddock

[maeder]

trying to build a binary ghc-7.1 under sparc solaris failed (late) as follows:

"inplace/bin/ghc-stage2"   -H32m -O -optc-mcpu=ultrasparc -opta-mcpu=ultrasparc -DNEW_GHC_LAYOUT   -hide-all-packages -i -iutils/haddock/src -iutils/haddock/dist/build -iutils/haddock/dist/build/autogen -Iutils/haddock/dist/build -Iutils/haddock/dist/build/autogen    -optP-DIN_GHC_TREE -optP-include -optPutils/haddock/dist/build/autogen/cabal_macros.h -package Cabal-1.10.0.0 -package array-0.3.0.2 -package base-4.3.0.0 -package containers-0.4.0.0 -package directory-1.1.0.0 -package filepath-1.2.0.0 -package ghc-7.0.0.20100924 -package pretty-1.0.1.2 -package xhtml-3000.2.0.1  -funbox-strict-fields -O2 -Wall -fwarn-tabs -XForeignFunctionInterface -XCPP -XPatternGuards -XDeriveDataTypeable -XScopedTypeVariables -XMagicHash -no-user-package-conf -rtsopts     -odir utils/haddock/dist/build -hidir utils/haddock/dist/build -stubdir utils/haddock/dist/build -hisuf hi -osuf  o -hcsuf hc -c utils/haddock/src/Haddock/GhcUtils.hs -o utils/haddock/dist/build/Haddock/GhcUtils.o
ghc-stage2: panic! (the 'impossible' happened)
  (GHC version 7.0.0.20100924 for sparc-sun-solaris2):
        System.Event.Array.unsafeWrite': bounds error, index 1, capacity 1

Please report this as a GHC bug:  http://www.haskell.org/ghc/reportabug

gmake[1]: *** [utils/haddock/dist/build/Haddock/GhcUtils.o] Error 1
gmake: *** [all] Error 2

[igloo]

I've reduced the problem to this:

$ cat q.hs

import GHC.Conc

main :: IO ()
main = do putStrLn "Q1"
          ensureIOManagerIsRunning
          putStrLn "Q2"

$ ../ghc-7.0.0.20100924/inplace/bin/ghc-stage1 --make q -threaded
[1 of 1] Compiling Main             ( q.hs, q.o )
Linking q ...
$ ./q
Q1
q: System.Event.Array.unsafeWrite': bounds error, index 1, capacity 1
$ 

[tibbe]

igloo says that HAVE_POLL is defined and HAVE_EPOLL is not, so this is likely a problem in the poll backend. If memory serves me right, the poll backend is also the only backend that uses System.Event.Array.

[tibbe]

The issue might be sparc specific as I couldn't reproduce this on OS X by forcing the I/O manager to use the poll backend. I will need an account on a sparc machine if I can't reproduce this on either my OS X or Linux machine.

[tibbe]

I cannot repro on my Linux machine either so I need access to a sparc machine.

[tibbe]

I'm having difficulty even building GHC on sparc. If someone could help me with the technical hurdles, it would be easier to to fix this bug. Here's the build error (on sparky.haskell.org):

$ ./configure --with-gcc=/opt/gcc-vanilla/4.1.2/bin/gcc --with-ghc=/opt/ghc/bin/ghc --prefix=/opt/ghc --with-gmp-includes=/opt/csw/include --with-gmp-libraries=/opt/csw/lib
$ make
"inplace/bin/hsc2hs" --cc=/opt/gcc-vanilla/4.1.2/bin/gcc --ld=/opt/gcc-vanilla/4.1.2/bin/gcc  --cflag=-fno-stack-protector -I/opt/csw/include  --cflag=-D__GLASGOW_HASKELL__=701 '--cflag=-fno-stack-protector' '--cflag=-Icompiler/../libffi/build/include' '--cflag=-Icompiler/stage2' '--cflag=-Icompiler/../libraries/base/cbits' '--cflag=-Icompiler/../libraries/base/include' '--cflag=-Icompiler/.' '--cflag=-Icompiler/parser' '--cflag=-Icompiler/utils' '--cflag=-DGHCI' '--cflag=-I/home/tibbe/ghc/master/libraries/bytestring/include' '--cflag=-I/home/tibbe/ghc/master/libraries/process/include' '--cflag=-I/home/tibbe/ghc/master/libraries/directory/include' '--cflag=-I/home/tibbe/ghc/master/libraries/unix/include' '--cflag=-I/home/tibbe/ghc/master/libraries/old-time/include' '--cflag=-I/home/tibbe/ghc/master/libraries/containers/include' '--cflag=-I/home/tibbe/ghc/master/libraries/array/include' '--cflag=-I/home/tibbe/ghc/master/libraries/base/include' '--cflag=-I/opt/csw/include' '--cflag=-I/home/tibbe/ghc/master/includes' '--cflag=-I/home/tibbe/ghc/master/libffi/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/template-haskell/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/hpc/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/bin-package-db/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/binary/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/bytestring/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/Cabal/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/process/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/pretty/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/directory/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/unix/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/old-time/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/old-locale/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/filepath/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/containers/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/array/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/base/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/libraries/integer-gmp/dist-install/build' '--lflag=-L/opt/csw/lib' '--lflag=-L/home/tibbe/ghc/master/libraries/ghc-prim/dist-install/build' '--lflag=-L/home/tibbe/ghc/master/rts/dist/build' '--lflag=-L/home/tibbe/ghc/master/libffi/dist-install/build' '--lflag=-lrt' '--lflag=-ldl' '--lflag=-lgmp' '--lflag=-lm' '--lflag=-lrt' '--lflag=-ldl'   compiler/ghci/LibFFI.hsc -o compiler/stage2/build/LibFFI.hs
ld.so.1: LibFFI_hsc_make: fatal: libgmp.so.10: open failed: No such file or directory
running compiler/stage2/build/LibFFI_hsc_make failed (exit code 9)
command was: compiler/stage2/build/LibFFI_hsc_make  >compiler/stage2/build/LibFFI.hs
make[1]: *** [compiler/stage2/build/LibFFI.hs] Error 1
make[1]: *** Deleting file `compiler/stage2/build/LibFFI.hs'
make: *** [all] Error 2

[igloo]

I believe I used

./configure --prefix=`pwd`/inst --with-gcc=/opt/gcc-vanilla/4.1.2/bin/gcc

[tibbe]

I cannot reproduce this with HEAD on sparky.haskell.org. Running igloo's test program q.hs multiple times works without a problem. I'll need more exact reproduction instructions. Which branch should I use?

[tibbe]

GHC also build fine for me. Here's the platform info:

$ uname -a
SunOS sparky 5.10 Generic_127111-11 sun4v sparc SUNW,SPARC-Enterprise-T5120

[maeder]

Ok, I'll try to build  http://new-www.haskell.org/ghc/dist/7.0.1-rc2/ghc-7.0.0.20101028-src.tar.bz2 to see if the problem is gone.

[maeder]

The problem is still present:

"inplace/bin/ghc-stage2"   -H32m -O -optc-mcpu=ultrasparc -opta-mcpu=ultrasparc -DNEW_GHC_LAYOUT   -hide-all-packages -i -iutils/haddock/src -iutils/haddock/dist/build -iutils/haddock/dist/build/autogen -Iutils/haddock/dist/build -Iutils/haddock/dist/build/autogen    -optP-DIN_GHC_TREE -optP-include -optPutils/haddock/dist/build/autogen/cabal_macros.h -package Cabal-1.10.0.0 -package array-0.3.0.2 -package base-4.3.0.0 -package containers-0.4.0.0 -package directory-1.1.0.0 -package filepath-1.2.0.0 -package ghc-7.0.0.20101028 -package pretty-1.0.1.2 -package xhtml-3000.2.0.1  -funbox-strict-fields -O2 -Wall -fwarn-tabs -XForeignFunctionInterface -XCPP -XPatternGuards -XDeriveDataTypeable -XScopedTypeVariables -XMagicHash -no-user-package-conf -rtsopts     -odir utils/haddock/dist/build -hidir utils/haddock/dist/build -stubdir utils/haddock/dist/build -hisuf hi -osuf  o -hcsuf hc -c utils/haddock/src/Haddock/GhcUtils.hs -o utils/haddock/dist/build/Haddock/GhcUtils.o
ghc-stage2: panic! (the 'impossible' happened)
  (GHC version 7.0.0.20101028 for sparc-sun-solaris2):
        System.Event.Array.unsafeWrite': bounds error, index 1, capacity 1

Please report this as a GHC bug:  http://www.haskell.org/ghc/reportabug

gmake[1]: *** [utils/haddock/dist/build/Haddock/GhcUtils.o] Error 1
gmake: *** [all] Error 2

Igloo's q-example still fails in the same way. (The -threaded flag is important.)

-bash-3.00$ uname -a
SunOS leo 5.10 Generic_142909-17 sun4u sparc SUNW,Sun-Fire-280R

[igloo]

I can reproduce it with HEAD on sparky.

[igloo]

I notice you have a mk/build.mk, with BuildFlavour = quickest. That's the only obvious difference between your build and mine.

[tibbe]

I've finally managed to reproduce the error. Looking into it now.

[tibbe]

Potential fix

[tibbe]

I believe I have a fix (see attached). It removes the problem in my tree, but given the build issues I've been having I'd like someone to verify. To test, just apply the patch to the base library, rebuild and run Ian's short example program (q.hs).

The bug is somewhere inside System.Event.Array.ensureCapacity'. I'm not sure exactly what's going on, but lots of printf-debugging later I have concluded that the firstPowerOf2, as used inside ensureCapacity' someone produces the wrong result, causing a too small array to get allocated. I cannot reproduce this on my x86 machine.

The attached patch has a better implementation of the firstPowerOf2 function.

[maeder]

Can someone tell me how to apply a (giant) darcs patch using gpatch on a plain source tree. I cannot find (or work out) how to do this.

[tibbe]

From inside libraries/base just run

darcs apply fix-4348.dpatch

[maeder]

I'm not working with HEAD via darcs!

-bash-3.00$ darcs apply fix-4348.dpatch 

darcs failed:  Unable to "darcs apply" here.

You need to be in a repository directory to run this command.

[tibbe]

The patch is quite small and you can view it using your favorite text editor. You can manually apply it to libraries/base/System/Event/Array.hs. It only changes one function and adjusts some imports.

[simonmar]

Perhaps the original version failed due to floating point inaccuracies? Anyway the new one is obviously better.

[tibbe]

Replying to simonmar:

Perhaps the original version failed due to floating point inaccuracies? Anyway the new one is obviously better.

Possibly, but I couldn't reproduce them even on the sparc machine. I extracted the firstPowerOf2 two function and tested it by feeding it a bunch of values. It produce the correct results. Very puzzling.

[igloo]

Applied to HEAD and 7.0; thanks!

[maeder]

Replying to tibbe:

The patch is quite small and you can view it using your favorite text editor. You can manually apply it to libraries/base/System/Event/Array.hs. It only changes one function and adjusts some imports.

What stopped you to attach this file! I'll now wait for a suitable (snapshot) distribution from http://www.haskell.org/ghc/dist/current/dist/

[tibbe]

Nothing in particular. I didn't think it would be that hard to get it from the GHC repo

 http://darcs.haskell.org/libraries/base/System/Event/Array.hs

[maeder]

Thanks for the link

[maeder]

Replacing this file indeed fixed the problem of building ghc-7.0.0.20101028

[maeder]

see #4475 for another problem