Re: [xmonad] Segfault on startup

6 Mar 2008

      almeidaraf:
...
On Thu, 6 Mar 2008 06:52:00 -0500, "Brent Yorgey"  wrote:
...
On Thu, Mar 6, 2008 at 1:41 AM, Don Stewart  wrote:
     roma:
     > * Rafael Almeida  [2008-03-06 02:57:16-0300]
     > > On Thu, Mar 6, 2008 at 2:36 AM, Don Stewart 
     wrote:
     > > > almeidaraf:
     > > >
     > > > > Hello,
     > > >  >
     > > >  > I've been using xmonad from darcs. Today I pulled the latest
     version
     > > >  > and, after compiling xmonad, I found it segfaults during the
     startup.
     > > >  > I'm running a debian 4.0 (etch) and I've installed X11 1.4.1
     on my
     > > >  > local user.
     > > >
     > > >  Almost certainly this is because you have a partially
     recompiled xmonad.
     > > >  Please do a full make clean, then try again.
     > > >
     > > >     runhaskell Setup.lhs clean
     > > >
     > > >  then rebuild.
     > > >
     > > >     http://haskell.org/haskellwiki/Xmonad/
     Frequently_asked_questions#Help.21_xmonad_just_segfaulted
     > > >
     > > >  Same for any config repo you use.
     > > >
     > >
     > > That doesn't seem the issue because I would always delete xmonad-
     new
     > > directory and recreate it using
     > >
     > > $ darcs get xmonad --to-patch=patchname xmonad-new
     > >
     > > in order to find out which patch broke it. Just in case I just
     did:
     > >
     > > $ darcs get http://code.haskell.org/xmonad --to-patch="Core.hs:
     add an
     > > Applicative" xmonad-new
     > >
     > > and compiled it. Sure enough, xmonad segfaulted on startup.
     >
     > As far as I see from your config, you also use xmonad contrib
     library.
     > Have you recompiled it as well?
Ah yes, that needs to be fully cleaned and recompiled.
And then confirm that without any config file it either is ok, or
     crashes.
Just to make sure everything is completely clear: every time you rebuild
xmonad, you must do a full clean and rebuild of xmonad-contrib.  Otherwise it
doesn't realize that a library it depends on has changed, and continues
referencing the old library, which may or may not cause a crash. =(
I'm pretty sure there's no way that simply adding an instance could cause a
crash, except for the recompilation problem explained above, or a bug in ghc.
-Brent
Oh yes. That was indeed the problem here. My bad :(. I was kinda thrown
away because I thought you'd never get a Segmentation Fault from a code
written in Haskell. But I supose that the error came from the C code on
X11, right? Shouldn't it return a nicer error instead of segfaulting? Or
is this behaviour expectable?
It actually came from GHC linking code that was compiled against a
different API. Due to cross module optimisations, its possible to
partially recompile some code, then modify some source, continue
compiling and link against an incorrect api.

Usually this results in undefined symbols, but occasionally you see
somehting like this.

If the source changes, via a darcs patch, always best to clean and
recompile. :)

-- Don