cmdSplitMeasure() must break notes in other Voices using toDurationList()

Well the attached:

Works in both 2.1-dev 5104373 (which has my commit) and 2.1-dev 1ffaed2 (the previous commit before my commit). So maybe I didn't break it. Maybe I need to type the score exactly according to the screenshot.

Ok, I figured out that the issue is not a *SegFault*, but rather is a Q_ASSERT failure, which only occurs when run on a *debug build*.

TDuration::TDuration(const Fraction& l, bool truncate, int maxDots, DurationType maxType)
      {
      setType(maxType); // use maxType to avoid testing all types if you know that l is smaller than a certain DurationType
      setDots(maxDots);
      truncateToFraction(l, maxDots);
      Q_ASSERT(truncate || (fraction() - l).numerator() == 0); // check for exact fit
      }

in that assert for my earlier attached example:

fraction() is a half note.
l is 11/16

Therefore fraction() - l doesn't evaluate to zero...so isn't an exact fit.

Note that the assert failure is still present on d4ea2418aacc2dfc100bad5993ddcd2588c83d6c which was the commit before maxDots changed from 4 into 3.

I am noting that on latest 2.1 commit, there is no assert failure when splitting before the dotted eight, or splitting before the first eight rest, but only when splitting before the sixteenth rest.

I am noting that a score with just the first measure will case the assert failure:

split184061-one-measure.mxl

Although if I remove the whole note in voice 2, then there is no longer an assert failure:

split184061-one-measure-one-voice.mscx

I am noting that the assert failure happens when there is a whole *note* rest in voice 2

split184061-one-measure-whole-note-voice-2.mscx

But doesn't happen when there is a whole *rest* in voice 2.

split184061-one-measure-whole-rest-voice-2.mscx

There is an actual bug, since the result in 2.1-nightly ae09a44:

splits the whole note into only a half note and a quarter note...which do not fill entire duration of whole note. Neither of the resulting notes fill their entire measure.

I'm changing variable name Fraction "rest" in TrackList::write to "remaining", because it appears to represent that reamining time in the measure being written to. "rest" is just confusing because a Rest is something different.

:-) ok...that is pretty humorous, I will admit.

I think I'll still change it to "remainder" for less ambiguity.

I'm trying to figure out why TrackList::write will break up rests according to toDurationList:

if (e->type() == Element::Type::REST || e->type() == Element::Type::REPEAT_MEASURE) {
        for (TDuration k : toDurationList(d, false)) {

But doesn't use toDurationList for breaking up chords, and instead just sets duration of note to d (the minimum of what is remaining in the measure and the duration of the note being split up:

        c->setDuration(d);

I think should use toDurationlist for both cases, to handle note duration which can't be easily split up into two smaller durations.

Two tests:

split184061-no-tie:

when split 11/16 through becomes:

and split184061-keep-tie:

when split 11/16 through becomes:

The strange thing is those test work from inside mscore, but the second one is not working from inside mtest because it is not copying that final quarter note when it saves the file for comparison. So I need to look into that. But other than that, I'm sharing my changes for others to comment before I make some more tests and submit PR:

https://github.com/musescore/MuseScore/compare/2.1...ericfont:184061-Sp…

The strange thing is those test work from inside mscore, but the second one is not working from inside mtest because it is not copying that final quarter note when it saves the file for comparison. So I need to look into that.

Turns out I had to add score->doLayout() to the test in order for that quarter note to actually become a segment in the initial score. I don't quite understand why that is necessary, to be honest.

I made a PR: https://github.com/musescore/MuseScore/pull/3108

I might add some more tests later. I do have a test which results in triple-dots...I hope that is ok:

When split becomes:

Actually I'd better readjust that last test so it doesn't have any rests, because the problem is split measure undoes the rest organization, but there is another feature request that says should instead try to preserve as much of repeats organization before and after the split.

Anyway I'd appreciate any suggestions about what strange behavior to test. I'm wondering if we allow to split measures where one or more instruments have a measure rest. I would think measure rests are unbreakable, so maybe we need to prohibit that...

I'm all in agreement that measure rests are extremely confusing way to represent the result of split measures. I think measure rests only should be used when measure's actual beats equals its time signature, which is not the case for newly-split measures. So I support that.

But in the latest pic I shared, you will notice that the split completely reorganized how the empty space is divided into rests. What I think is if there are already rest elements (other than Measure Rest), then each of those rest elements should be preserved in same manner that they were before the split if possible (and if that isn't possible, for instance if a rest was straddling the tick of the split, then that rest should be broken up). But no combining of rests. There is always the "Simplify Durations" function which user could manually apply after performing the split, which could merge rests according to the time Sig's subdivision if desired.