GHC 6.4.3 is stalled

[Simon Marlow]

I just realised I sent the following only to cvs-ghc at haskell.org, not
glasgow-hasell-users:

Simon Marlow wrote:
> Just a quick note to say thanks for all the offers of help on 6.4.3 - 
> it's very encouraging to see so many people willing to offer their time 
> and machines.
> 
> The state of play so far:
> 
> On FreeBSD, reports suggest that there's are one or possibly two 
> problems exposed by the threaded RTS.  The timeout program in the 
> testsuite sometimes hangs, and GHC itself may hang when linked against 
> the threaded RTS.  This looks like a tricky one, so I'm going to try to 
> look into it myself.  I've installed FreeBSD 6.1 on an old laptop and 
> I'm currently building 6.4.2 on it.
> 
> On MacOS X / PPC, we have a crashing stage2 when linked against the 
> threaded RTS.  Gregory Wright is trying to reproduce the problem on a 
> smaller scale (with something other than GHC).
> 
> The "timeout hanging" problem apparently happens on MacOS X/Intel too 
> (according to Joel Reymont).  If I find the cause on FreeBSD, we can 
> test the fix on MacOS X/Intel.

I have an update on FreeBSD.  I built the CVS 6.4 branch on a fresh install of
FreeBSD, and ran the testsuite:

OVERALL SUMMARY for test run started at Thu  3 Aug 2006 09:42:30 BST
     1365 total tests, which gave rise to
     4157 test cases, of which
        0 caused framework failures
      583 were skipped

     3505 expected passes
       46 expected failures
        2 unexpected passes
       21 unexpected failures

Unexpected passes:
    cholewo-eval(optasm)
    ffi009(optasm)

Unexpected failures:
    arith011(opt)
    barton-mangler-bug(normal,opt,ghci,threaded)
    cabal01(normal)
    galois_raytrace(ghci)
    ioref001(normal,threaded)
    joao-circular(normal,opt,optasm,threaded)
    newtype(opt)
    queryfdoption01(normal,opt,optasm,threaded)
    rn.prog006(normal)
    seward-space-leak(ghci)
    signals002(threaded)


Some of those failures are probably due to running out of memory on this laptop
(which only has 128M + 256M swap, I should have given it more swap).  But I
don't see any evidence of hanging timeout processes and stage2 apparently works
fine.

That means the problem only shows up on a dual proc/dual core (Gregory reported
seeing this on a dual proc build of FreeBSD).  Greg - can you give me SSH access
to your box, and I'll take a look?

Cheers,
	Simon