Comment 1 by bcfabien:

Russ,
DBL_EPSILON is used frequently in numerical computations.
Here are two examples.
1) If f1 and f2 are O(1) then consider the following test:
 if math.Fabs(f2-f1) <= DBL_EPSILON*(1.0 + math.Fabs(f1))
if this test is true then relative error between f1 and f2 is negligible.
2) math.Sqrt(DBL_EPSILON) is related to the "optimal" step size used in forward
difference approximation of derivatives.

Comment 2:

I'm happy to defer to the real numericists.  I can see how it would be useful
though I'm skeptical that most people use it correctly.  
EpsilonFloat32 and EpsilonFloat64?

Comment 3 by bcfabien:

I only use float32 when I want to illustrate the deficiencies of single precision.
So, I would vote for EpsilonFloat64.
However, some programmers may wish for EpsilonFloat32 as well.
Thanks.

Comment 4 by joerg.sonnenberger:

ISO C99 has nextafter and co for this purpose. Epsilon itself doesn't make much sense as
it depends on the value (think denormalized numbers).

Comment 5:

Go has Nextafter too.  Good enough.

Status changed to WorkingAsIntended.