fast fitting: simplify loop logic

[woutdenolf]

We have been using FastXRFLinearFit with a patched OutputBuffer for a while in ewoks to have full control over the output but we had to disable refitting since it's a one-way street. In addition this code was insanely complex. So I decided to bite the bullet and use ClassMcaTheory with ConcentrationsTool directly.

I'm in the process of dissecting the FastXRFLinearFit code and several merge requests will follow.

This MR tackles only the loop itself. In a follow-up I will tackle the bad pixel stuff. There is another MR on the xmin/xmax limits.

def original_while_loop(nFreeBkg, nFree):
    iIter = 1
    nIter = 2 * (nFree - nFreeBkg) + iIter
    negativePresent = True

    nfits = 0
    print(nIter)

    while negativePresent:
        if iIter > nIter:
            negativePresent = False
            print("max")
            continue
        nfits += 1  # refit
        iIter += 1

    print(nfits)


def new_for_loop_with_break(nFreeBkg, nFree):
    nPeakAreas = nFree - nFreeBkg
    nrefitMax = 2 * nPeakAreas + 1

    nfits = 0
    print(nrefitMax)

    for refitIter in range(1, nrefitMax + 2):
        if refitIter > nrefitMax:
            print("max")
            break
        nfits += 1  # refit

    print(nfits)


if __name__ == "__main__":
    original_while_loop(0, 5)
    print()
    new_for_loop_with_break(0, 5)
    print()
    original_while_loop(5, 5)
    print()
    new_for_loop_with_break(5, 5)

[woutdenolf]

What is the reason for the maximum number of refits to be 2 * nPeakAreas + 1?

[vasole]

If you have repeated the fit for all the elements twice, you surely are in an endless loop. Therefore a limit based on the number of fitted peak areas.

[vasole]

We have been using FastXRFLinearFit with a patched OutputBuffer for a while in ewoks to have full control over the output but we had to disable refitting since it's a one-way street. In addition this code was insanely complex. So I decided to bite the bullet and use ClassMcaTheory with ConcentrationsTool directly.

The original/legacy implementation without OutputBuffer was simpler and on those lines.

[vasole]

The change on the limits is to be reverted. McaAdvancedFitBatch and FastXRFLinearFit follow the same philosophy: if no limits are passed, the limits in the fit configuration are used.

use_limits was implemented on user request to avoid having irreproducible interactive sessions because of not selecting exactly the same zoom region.