Beam broken segment: wrong position

Context: commit 8e8ddd7bb (22-11-12)

Step: load or re-created the attached sample score

Result: the broken beam for the first 16th is on the wrong side:

Notes:
This is related to, but possibly more general than, issues #14869: [trunk] beam on wrong side for semi-quavers in 6/8 (or similar) time and #15779: 32nd note flag points wrong way in sub-beam. Both were solved at some moment but at least the second is back.

Werner noted once that there are occasions when either beam direction might be needed according to the intended rhythm and a way should be provided to obtain either on request. This would be great, but in the meantime, the most common case should give the most commonly 'correct' result.

Analysis:
The code involved is in file libmscore/beam.cpp, function Beam::layout2(), lines 1780 and following:
   if (a.reduced().denominator() < b.reduced().denominator())
     len = -len;
It attempts to compare the durations of the chord with the broken segment, of the previous chord and of the following chord.

This approach fails to distinguish between the groups marked 1 and 2 in the image above: they have the same duration pattern but require different positions for the broken beam.

A better approach could to determine the position of the beam group in the measure by replacing the above lines with (all steps shown in full; code can be optimized):
   Measure* meas = cr1->measure();
   int measTick = meas->tick();
   ChordRest* cr1 = crl[c1-1];
   ChordRest* cr2 = crl[c1];
   int tick1 = cr1->tick() - measTick;
   int tick2 = cr2->tick() - measTick;
   int ticks = ab.ticks();
   int mod1 = tick1 % ticks;
   int mod2 = tick2 % ticks;
   if (mod1 < mod2)
     len = -len;
if the previous chord is on a 'more important' division of the measure than the target chord, the beam is turned to the left. The example above becomes:

If the approach seems correct, I can provide a pull request with a more compact code.

Thanks,

M.