bug in profiling compiler version

[Kevin Glynn]

Hi Uwe,

I'm sorry about the delay replying to this message.  I've investigated
your problem and its another of those "Hmmm,  why doesn't this break
all the time things".  I'll check a fix into the CVS sources.

In the meantime I have two possible work arounds:

The first is to change the definition of StateTrans in StateTrans.hs
from a newtype to a datatype. Unfortunatley this may distort your
profiles, so .....

The second is a hack, you'd better email me if you want it.

Incidentally,  you will need to add the -prof option to the linking
phases of pplc and testNtree in the Makefile.


regards
Kevin
 



Uwe Schmidt writes:
 > hi ghc group,
 > 
 > I'm developing a small compiler for an experimental
 > language for image generation, and i tried the profiling
 > option to analyse where all the memory is used, and
 > then the impossible occured:
 > 
 > here is the listing:
 > 
 > ----------------------
 > 
 > si(4):~:39> cd tmp
 > si(4):~/tmp:40> tar xvzf bug.tgz
 > pl0/AbstractSyntax.hs
 > pl0/Assemble.hs
 > pl0/BuildinFunctions.hs
 > pl0/BuiltinFunctions.hs
 > pl0/CodeGeneration.hs
 > pl0/ControlUnit.hs
 > pl0/Error.hs
 > pl0/GlobalState.hs
 > pl0/Instructions.hs
 > pl0/Lexer.hs
 > pl0/Loader.hs
 > pl0/MachineArchitecture.hs
 > pl0/MachineStructure.hs
 > pl0/Main.hs
 > pl0/Matrix.hs
 > pl0/MicroCode.hs
 > pl0/NTree.hs
 > pl0/OPCode.hs
 > pl0/PPLInterpreter.hs
 > pl0/Parser.hs
 > pl0/Picture.hs
 > pl0/Ppm.hs
 > pl0/SemanticAnalysis.hs
 > pl0/ShowAbstractSyntaxTree.hs
 > pl0/ShowAttrTree.hs
 > pl0/ShowCode.hs
 > pl0/ShowJavaCode.hs
 > pl0/ShowMS.hs
 > pl0/StateTrans.hs
 > pl0/Symbol.hs
 > pl0/TestNTree.hs
 > pl0/t.hs
 > pl0/Makefile
 > si(4):~/tmp:41> cd pl0/
 > si(4):~/tmp/pl0:42> make ControlUnit.o
 > ghc -prof -auto-all -K20m -H100m -O -c Instructions.hs
 > ghc -prof -auto-all -K20m -H100m -O -c Matrix.hs
 > ghc -prof -auto-all -K20m -H100m -O -c Picture.hs
 > ghc -prof -auto-all -K20m -H100m -O -c MachineArchitecture.hs
 > ghc -prof -auto-all -K20m -H100m -O -c StateTrans.hs
 > ghc -prof -auto-all -K20m -H100m -O -c Error.hs
 > ghc -prof -auto-all -K20m -H100m -O -c ShowCode.hs
 > ghc -prof -auto-all -K20m -H100m -O -c ShowMS.hs
 > ghc -prof -auto-all -K20m -H100m -O -c MicroCode.hs
 > ghc -prof -auto-all -K20m -H100m -O -c Ppm.hs
 > ghc -prof -auto-all -K20m -H100m -O -c OPCode.hs
 > ghc -prof -auto-all -K20m -H100m -O -c ControlUnit.hs
 >  
 > panic! (the `impossible' happened):
 >         CPR Analysis tried to take the lub of a function and a tuple
 >  
 > Please report it as a compiler bug to glasgow-haskell-bugs at haskell.org.
 >  
 >  
 > make: *** [ControlUnit.o] Error 1
 > si(4):~/tmp/pl0:43> ghc --version
 > The Glorious Glasgow Haskell Compilation System, version 4.08.2
 > si(4):~/tmp/pl0:44> cat /proc/version
 > Linux version 2.2.16 (root at sparbier) (gcc version 2.95.2 19991024
 > (release)) #4 Thu Oct 5 18:40:46 CEST 2000
 > si(4):~/tmp/pl0:45>      
 > 
 > -------------------------------------------------------------------
 > 
 > the error does not occur without the profiling option,
 > but it also occurs without -O and with -O2 instead of -O
 > 
 > 
 > I hope, it's not too hard to find it
 > 
 >   uwe schmidt
 > 
 > -- 
 > ____________   ______
 > \   \~~~\   \  \    /   Fachhochschule Wedel  | Prof.Dr. Uwe Schmidt
 >  \   \   \   \  \  /   Feldstrasse 143        | 
 >   \   \== \   \==\/   D-22880 Wedel           | Tel  : +49-4103-8048-45
 >    \   \   \   \     Fachbereich              | Fax  : +49-4103-8048-39
 >     \  /    \  /    Wirtschaftsinformatik     | EMail: uwe at fh-wedel.de
 >      \/      \/    "http://www.fh-wedel.de/~si/index.html