Profiling bug in 5.04.2 ?

[Julian Seward]

Greetings, y'all.

I've got a small 17000 line Haskell program which works fine
under the following sets of flags (also uses -package data 
-package text), which are not shown:

ghc -o main --make Main.hs      -fasm-x86

ghc -o main --make Main.hs   -O -fasm-x86

ghc -o main --make Main.hs   -O -fvia-C

However, when compiled for profiling ...

ghc -o main --make Main.hs   -O -fvia-C -prof -auto-all

the resulting binary fails, after a while, with this:
   Fail: Prelude.foldl1: empty list
This is without any +RTS -p whatever options or any such.

I tried re-running with different heap sizes (various 
combinations of -H100m, -A40m) and it makes no difference.

This is on Red Hat 7.3, glibc-2.2.5, the infamous 
gcc version 2.96 20000731 (Red Hat Linux 7.3 2.96-113)
and ghc-5.04.2 I built from source myself via the usual 2-stage
bootstrap and it has worked otherwise flawlessly for months.

Any ideas?  Are there known profiling bugs in 5.04.2 ?

I guess I will try ghc-6.0 and see what happens.

J