bug on OSX "only" in some cases in new fast GAUSSFIT

[alaingdl]

No problem up to now on Linux, 2 very different OSX gave the same error (b1)

pro TEST_BUG_GAUSSFIT

a = [4.0, 1.0, 2.0, 1.0]

n=100
x = (FINDGEN(n)-(n/2))/4
z = (x -a[1])/a[2]
y = a[0]*exp(-z^2/2)
y=y+0.01*randomn(123, n)

yf=gaussfit(x, y)

plot, x, y
oplot, x, yf, col=color2color(!color.green)

yf1=gaussfit(x, y, b1)
oplot, x, yf1, col=color2color(!color.red)

b=a+0.1
yf2=gaussfit(x, y, c, est=b, nt=4)
oplot, x, yf2, col=color2color(!color.orange)

print, b1
print, c

stop
end

[alaingdl]

I wrote some tests covering more tests for GAUSSFIT() and the situation is not good !
-- some tests (when called without estimated values) are broken on OSX (but not on Linux)
-- few basic cases are now broken in Linux too

The good point is the code is very fast when it is stable !

[alaingdl]

Since now we do have a "test_gaussfit.pro" and a "bench_gaussfit.pro" in Git, and I correct the code to take into account the case when the step in X is not unity (1), and I also improved the initial estimation of the a4/a5/a6 and the result is in #1671 I will close this issue.

Please open a new issue if any trouble with GAUSSFIT()

Sorry but few keywords now are not avaialble in the fast GAUSSFIT, you can use the old one ("gaussfit_mpfit.pro" in src/pro) for that.

On larges tests I made, the new implementation is as fast and as accurate than the IDL one ...