fminbnd test failure

[Ben Abbott]

I'm seeing a new failure for fminbnd, which I assume is due to the change below.

        http://hg.savannah.gnu.org/hgweb/octave/rev/b4e5dcf023c9

Perhaps the tolerance should be sqrt(eps)?

octave:1> test fminbnd 
 ***** assert (fminbnd (@cos, pi/2, 3*pi/2, opt0), pi, 10*eps)
!!!!! test failed
assert (fminbnd (@cos, pi / 2, 3 * pi / 2, opt0),pi,10 * eps) expected
3.1416
but got
3.1416
maximum absolute error 1.01056e-08 exceeds tolerance 2.22045e-15
shared variables {
 opt0 =
 {
   TolX = 0
 }

}

Ben