[mercury-users] [petdr at miscrit.be: Report on the ICFP 2000 prog contest]

[Thomas Conway]

On Wed, Sep 06, 2000 at 09:12:50PM EST, schachte at cs.mu.OZ.AU wrote:
> I'll be interested to hear how quickly (how many questions) the
> declarative debugger finds that needle in the haystack.

For us, doing it by hand (I used mdb, I don't know about the others),
analysis of the image told us that a certain kind if insideness test
was failing where it should succeed. Conveniently, we were able to put
a breakpoint on a piece of code that we knew would only call it in a way
that should succeed. In general, we had to debug by careful analysis
of the image and the code rather than the usual "watch what it's doing"
approach.

There was another bug that Mark and I had in some anti-surface-acne
code. When you're computing the lighting, you need to make sure the
object whose surface you're lighting doesn't obscure itself due to
numerical roundoff. So Mark and I added a "unique" number to each
object, and changed the code to use that so that objects couldn't
obscure themselves. This worked beautifully except that we noticed
that some reflections were missing from one of the images. Very
mysterious! We puzzled over it for ages until we realized that in
GML you can create an object an then use it multiple times, transformed
in different ways. The two spheres in the image were only the one
sphere, so interections of rays from one into the other were being
ignored. Very subtle. I don't know how you'd use a declarative debugger
to track that one down.

-- 
 Thomas Conway              Unicode for terrorists: U+0001 300C
 <conway at cs.mu.oz.au>         "Tie his hands behind his back"
 Mercurian )O+  
--------------------------------------------------------------------------
mercury-users mailing list
post:  mercury-users at cs.mu.oz.au
administrative address: owner-mercury-users at cs.mu.oz.au
unsubscribe: Address: mercury-users-request at cs.mu.oz.au Message: unsubscribe
subscribe:   Address: mercury-users-request at cs.mu.oz.au Message: subscribe
--------------------------------------------------------------------------