5.004_02: Complex/Trig: update
The following patches do not fix actual grave errors but they do:
- make the code more robust (more discontinuities catched)
(e.g. atan(-i), atanh(-1))
- make the results agree on signs and/or conjugate forms with the
results MATLAB gives: the results were already correct thanks to
the periodicity of trig funcs but now they are also consistent.
(e.g. acos(x) did have an unnecessary discontinuity at x = 0)
- for some pure real arguments short-circuit the calculation
to avoid rounding errors (which make epsilons appear where
clear zeros should reign)
Tested on NetBSD 1.2G i686, Linux 2.0.25 i686, Digital UNIX 4.0 EV56.
p5p-msgid:
199708081842.VAA31214@alpha.hut.fi