Hello,
	I recently downloaded the sys/lib/libm/erf.c and sys/lib/libm/mathimpl.h
files from NetBSD in order to implement the complementary error function
for an XFCN to be used in Spinnaker Plus on the Mac. In doing so, I found
that for inputs of x < -1.25 the result was not what I expected. By changing
one line of the code from two-r to two+r, I was able to get our expected
results.
	My findings should be tempered by the fact that I did not take
the time to implement the exp__D function, instead replacing:

	exp__D(s, y)

with:   exp(s)

	I'm not sure in exactly what way this may have affected my result,
but I believe my findings were independent of the changes to exp__D, because
I tried many variations of input to my replacement of exp__D, and they
did not affect the results of erfc(x < -1.25) that I was looking at.

Andy Arenson
arenson@rice.edu

------------------------------------------------------------------------------