Verify GHC 6.10/x86_64 optimizer fix

[GoogleCodeExporter]

Steps to reproduce the problem:
1. run an XMonad.Prompt prompt  

What is the expected output? What do you see instead?
A prompt is supposed to pop up, but nothing happens, and xmonad stops
responding to any input

Attached it xmonad's stdout

What version of the product are you using? On what operating system?
current darcs of: x11, core and contrib

ghc version 6.10.0.20081007

xorg server 1.5.2, libx11 1.1.5 (Archlinux testing repo)

Please provide any additional information below.

A minimal config; freezes on M-x:

> import XMonad (xmonad,defaultConfig)
> import XMonad.Prompt (defaultXPConfig)
> import XMonad.Prompt.XMonad (xmonadPrompt)
> import XMonad.Util.EZConfig (additionalKeysP)
> main = xmonad $ defaultConfig `additionalKeysP` [("M-x", xmonadPrompt
defaultXPConfig)]

If this error is due to a module from XMonadContrib, please tag this issue
with 'Component-Contrib' below.
Component-Contrib

Original issue reported on code.google.com by vogt.a...@gmail.com on 20 Oct 2008 at 5:22

Attachments:

• err2

[GoogleCodeExporter]

Exactly the same problem here.

Original comment by jan.h....@gmail.com on 11 Nov 2008 at 4:18

[GoogleCodeExporter]

Similar results with 6.10.1, also with gridSelect

See those similar, but not exactly the same x11 errors attached:

Original comment by vogt.a...@gmail.com on 13 Nov 2008 at 3:26

Attachments:

• error

[GoogleCodeExporter]

Original comment by SpencerJ...@gmail.com on 23 Nov 2008 at 11:03

• Added labels: Milestone-Release0.9, Priority-Critical
• Removed labels: Priority-Medium

[GoogleCodeExporter]

Original comment by SpencerJ...@gmail.com on 23 Nov 2008 at 11:06

[GoogleCodeExporter]

What architecture are you guys on? As it happens, I am running the release GHC 
with
darcs XMonad/XMC, but I don't notice any issues with Prompt - but I'm on an 
i386 Ubuntu.

Original comment by gwe...@gmail.com on 23 Nov 2008 at 11:27

[GoogleCodeExporter]

This is on x86_64

Original comment by vogt.a...@gmail.com on 24 Nov 2008 at 2:52

[GoogleCodeExporter]

Since we see this error on x86_64, but not i386, it is logical to assume that 
the
issue is most likely in our FFI code.  The next step is to audit all the X11 
calls
that prompt makes and check that the proper sized types are used.

Original comment by SpencerJ...@gmail.com on 5 Dec 2008 at 5:00

[GoogleCodeExporter]

With 5 Dec xm darcs (with xft), X11-1.4.5, xorg-server-1.5.2 on i686 and x86_64 
machines:

simpleTabbed works on i686, but on x86_64 allocates space for tabs and draws 
nothing,
does not freeze. (i.e. a gap appears with root visible) 
http://hpaste.org/12713#a4 --simpleTabbed stdError output

shellPrompt works on i686, does nothing then freezes on x86_64. 
http://hpaste.org/12713#a2 --shellPrompt stdError output

Original comment by wirtwo...@gmail.com on 5 Dec 2008 at 4:38

[GoogleCodeExporter]

Recompiled without xft and still get same errors (simpleTabbed shellPrompt) on
x86_64/6.10. Attached is ghc-pkg list and output from sjanssen's test code.

Original comment by wirtwo...@gmail.com on 5 Dec 2008 at 8:29

Attachments:

• xftTestLog

[GoogleCodeExporter]

Thanks for all the excellent test data.  All tests so far have an error similar 
to:

> xmonad: X11 error: BadMatch (invalid parameter attributes), request code=1, 
error
code=8

Which indicates an invalid parameter (BadMatch) on a XCreateWindow call (X11 
opcode
1).  Naturally all subsequent X calls operating on that window will fail, 
leading to
a failure to render windows and, in the case of Prompt which gathers input from 
the
window, lock ups.

We have also seen that superficially similar programs have different results,
sometimes even varying from machine to machine.

All of these results point to one of two things: incorrect foreign import
declarations for createNewWindow, or correct foreign declarations but incorrect 
code
generation by GHC.

The next step is to come up with a minimal test program and ask for help from 
GHC HQ.

Original comment by SpencerJ...@gmail.com on 6 Dec 2008 at 7:40

[GoogleCodeExporter]

xmobar-0.9 and xmobar-darcs also trigger similar error when xmobar is run from 
cli or
spawned in xmonad.hs. These were xmobars compiled with ghc-6.10.1 on two 
different
64-bit gentoo systems. Reverting to ghc-6.8.3 and rebuilding haskell bits makes
prompt/showWS/decorations/xmobar work again.

Original comment by wirtwo...@gmail.com on 26 Dec 2008 at 2:10

[GoogleCodeExporter]

This seems to be a GHC optimization bug.  As a workaround, try this in 
xmonad-contrib:

> runghc Setup clean
> runghc Setup configure --disable-optimization # and any other flags you 
usually pass
> runghc Setup build
> runghc Setup install
> xmonad --recompile

Please advise whether this solves the issue or not.

Original comment by SpencerJ...@gmail.com on 8 Jan 2009 at 2:35

[GoogleCodeExporter]

This fixes the issue for me, thanks.

Original comment by frederik...@gmail.com on 8 Jan 2009 at 10:34

[GoogleCodeExporter]

This fixes it for me too

Original comment by vogt.a...@gmail.com on 8 Jan 2009 at 11:39

[GoogleCodeExporter]

This has been bothering me for some time in my Gentoo install.

That fix worked for me too, so I attach a modified ebuild for Gentoo which 
includes
that flag and a warning with instructions on how to proceed after recompiling 
it. I
didn't know how to contact the haskell overlay team, I hope someone sees the 
ebuild
here. I hope it's useful to someone. It seems to have worked for me at least.

Original comment by yobonobo...@gmail.com on 9 Jan 2009 at 5:50

Attachments:

• xmonad-contrib-darcs-0.ebuild

[GoogleCodeExporter]

worked here as well, darcs xmonad-contrib, ghc 6.10 on arch linux.

Original comment by sean.esc...@gmail.com on 12 Jan 2009 at 5:32

[GoogleCodeExporter]

Using CentOS 5.1 with GHC 6.10.1 x86_64. Same issue. Disabling optimization 
fixed
this for me as well.

Original comment by coreyoco...@gmail.com on 16 Jan 2009 at 5:53

[GoogleCodeExporter]

We need to construct a test case for the GHC team, and submit a bug report 
here, 

http://hackage.haskell.org/trac/ghc/newticket?type=bug

Original comment by don...@gmail.com on 16 Jan 2009 at 5:56

[GoogleCodeExporter]

Now we wait for GHC HQ.

http://hackage.haskell.org/trac/ghc/ticket/2961

Original comment by SpencerJ...@gmail.com on 16 Jan 2009 at 11:43

• Added labels: Priority-Low
• Removed labels: Milestone-Release0.9, Priority-Critical

[GoogleCodeExporter]

The bug has been fixed in GHC head.  Now we wait for the next GHC release to 
verify.

Original comment by SpencerJ...@gmail.com on 26 Jan 2009 at 9:04

• Changed title: Verify GHC 6.10/x86_64 optimizer fix
• Changed state: Started

[GoogleCodeExporter]

I can't reproduce this with ghc 6.10.2

Original comment by vogt.a...@gmail.com on 9 Apr 2009 at 1:56

[GoogleCodeExporter]

Original comment by SpencerJ...@gmail.com on 9 Apr 2009 at 8:04

• Changed state: Verified