Need help with debugging "Impossible case alternative"

[Jan Stolarek]

Once again thanks for explanation! I found the problem, so now I only have a handful of failing 
tests, which hopefully will be easy to fix.

> That's very dangerous!  You *only* want to do this for the values returned
> by your primops, not for anything else.  Maybe that's the only way that
> trueVal, falseVal are used, but I'd change their name to trueValInt,
> falseValInt, and add a careful comment to explain that they return an
> expression of type Int# not Bool.
OK, I will document that.

Janek