assignment problem with doubles: getting -NaN unexpectedly

[Monty]

Hi Karl,

I've run into the same problem and done some research.  At this point, I'm
trying to establish if there are two seperate FP issues running around or only
one.

One bug is 100% confirmed and understood.  It seems that glibc (both 2.0 and 2.1) has several math functions that add entries to the FP stack blindly (and egcs doesn't realize this when laying out stack usage).  The result is an FP stack overflow that may not be immediately apparent.  GCC <= 2.7.x does not exhibit the problem due to inlining being disabled.  

So, if -D__NO_MATH_INLINES magically fixes your code, we've found your bug (and I'm only chasing one problem).  If this does not correct the problem, then there's a nother more sinister bug left to catch.  Let me know.  My ass is in this same fire ;-)

Monty