Richard Fateman wrote:
> That seems to describe what happens in this case.
I tried to be very careful when making this change, but obviously I
wasn't careful enough.
I can see now what is happening, sort of. For the case sqrt(2)/2, TMS
computes the product of 1/2 * sqrt(2), which gets simplified to 1/sqrt(2).
But for sqrt(2)/-2, TMS gets (-1/2)*sqrt(2) and the code does something
different because it doesn't recognize -1/2 as being essentially
equivalent to 1/2.
Note also that 1/2*sqrt(2) is handled differently from sqrt(2)/2. In
the former case, TMS sees sqrt(2)*(1/2).
All of this happens because TMS gets individual factors in different
orders and thus gets different results.
Yuck.
Ray