[Haskell] modern language design, stone age tools

Robert Ennals Robert.Ennals at cl.cam.ac.uk
Wed Jun 23 04:27:26 EDT 2004


I wrote such a debugger as part of my PhD work. It is called "hsdebug" and you 
can read the Haskell Workshop paper on it here:

http://www.cambridge.intel-research.net/~rennals/hw2003.pdf

Unfortunately, HsDebug depends on Optimistic Evaluation, and it seems unlikely 
that Optimistic Evaluation will appear in a release version of GHC (too hard 
to maintain, and I've gone off to work for Intel Research).

HsDebug is a GDB-style debugger which takes advantage of the fact that 
Optimistic Evaluation makes programs evaluate in a largely-strict evaluation 
order. It also uses a trick called "transient tail frames" to allow tail calls 
to be visible in stack traces.

HsDebug can debug any GHC-compilable Haskell program, including GHC itself. 


-Rob


> \begin{gripe}
> 
> Seeing as Haskell is apparently such a popular language these days,
> I don't suppose a working debugger would be too much to ask for, would it?
> Or even just a decent call stack trace when a program terminates with an exception?
> 
> In case you're wondering, yes I have already tried using Hat and Buddha.
> But I'm trying to debug a real application, not a toy one, and neither Hat
> nor Buddha support enough of Haskell.  Buddha doesn't even come close,
> as far as I can tell, and Hat lacks support for library modules such as Data.Word.
> 
> I've also tried using ghc's profiling to get stack traces, but that doesn't
> work very well either.  Firstly, the stack traces that you get only include
> function names, not line numbers.  Secondly, the stack traces are often
> incomplete, for reasons that I don't fully understand.  One likely cause is
> tail call optimization.  (Is there any option to switch off tail call optimization?
> I didn't see any such option in the ghc man page.)
> Thirdly, profiling seems to be incompatible with the use of ghci; there
> doesn't seem to be any easy way to build a workspace so that you can
> get stack traces and use ghci in that workspace at the same time.
> And ghc is slow enough (even on a 3.2GHz Pentium 4) that recompiling
> the whole workspace is highly unpalettable.
> 
> There's a hell of a lot of languages out there that _do_ support decent
> stack traces when an exception is thrown.  
> 
> What's the point of using a high-level functional language if it means
> you're stuck with poor library support and/or stone-age tools?
> 
> \end{gripe}
> 
> -- 
> Fergus J. Henderson                 |  "I have always known that the pursuit
> Galois Connections, Inc.            |  of excellence is a lethal habit"
> Phone: +1 503 626 6616              |     -- the last words of T. S. Garp.
> _______________________________________________
> Haskell mailing list
> Haskell at haskell.org
> http://www.haskell.org/mailman/listinfo/haskell
> 




More information about the Haskell mailing list