Avoid failed assertion in stem tremolo code

Avoid failed assertion in stem tremolo code

Example code triggering the problem:

    \relative { a32 8..:32 }

[Keith, 2019-04-03 08:16:22]

If I understand, the example
  \relative { a32 8..:32 }
fails the assertion in the call to center() on line 182, because the interval
'ph' is empty.

There are several problems in the existing code, that would seem to prevent it
from working.  The proposed change deviates further from what seems to be the
intent of the code, so I suggest trapping for an empty interval explicitly.

https://codereview.appspot.com/572550043/diff/560590043/lily/stem-tremolo.cc
File lily/stem-tremolo.cc (left):

https://codereview.appspot.com/572550043/diff/560590043/lily/stem-tremolo.cc#...
lily/stem-tremolo.cc:156: Stem_tremolo::pure_height (SCM smob, SCM, SCM)
Stem_tremolo::pure_height (SCM smob, SCM /*start*/, SCM /*end*/)
The duty of this function is to return the vertical extent of the tremolo, for
purposes of figuring the length required to set a line of music that begins at
column 'start' and ends at column 'end'.

https://codereview.appspot.com/572550043/diff/560590043/lily/stem-tremolo.cc#...
lily/stem-tremolo.cc:175: Interval ph = stem->pure_y_extent (stem, 0, INT_MAX);
The code clearly assumed that 'ph' would never be empty.  
In the given example, it certainly should not be empty, so there is a problem in
stem.cc.  

If in future stem->pure_y_extent() is repaired, and validly returns an empty
interval, this function should behave as it does with no stem at all:
  if (is_empty(ph)) return s1.extent(Y_AXIS);

https://codereview.appspot.com/572550043/diff/560590043/lily/stem-tremolo.cc#...
lily/stem-tremolo.cc:177: ph[-dir] = si.shortest_y_;
shortest_y is the endpoint of the stem (including space for tremolo) when it is
compressed as short as it can be.  So the operation above would normally be a
shortening of the interval.  
Doing instead add_point() would defeat its purpose.

https://codereview.appspot.com/572550043/diff/560590043/lily/stem-tremolo.cc#...
lily/stem-tremolo.cc:182: ph = ph - ph.center ();
There is the obvious problem that line 182 destroys the effort of line 181.   It
reports every stem-tremolo to be centered on the central staff-line, which makes
line-spacing less accurate but could easily go un-noticed.

[Keith, 2019-04-03 17:10:50]

https://codereview.appspot.com/572550043/diff/560590043/lily/stem-tremolo.cc
File lily/stem-tremolo.cc (left):

https://codereview.appspot.com/572550043/diff/560590043/lily/stem-tremolo.cc#...
lily/stem-tremolo.cc:175: Interval ph = stem->pure_y_extent (stem, 0, INT_MAX);
My suggested change to return s1.extent will probably cause green-highlights in
the regression tests, because the highlighting function compares the numerical
/extents/ of the graphical objects.  If the spacing of the black-printed glyphs
remains good, everything should be fine.

[dak, 2019-04-08 16:14:53]

Incorporate Keith's feedback

[dak, 2019-04-08 16:17:16]

Thanks for looking at this, Keith!  I've tried to make a version that does the
right thing with regard to the assertion failure.  That does not address the
root cause of the problem, nor does it fix the centering thing you pointed out. 
Quite lacklustre.

https://codereview.appspot.com/572550043/diff/560590043/lily/stem-tremolo.cc
File lily/stem-tremolo.cc (left):

https://codereview.appspot.com/572550043/diff/560590043/lily/stem-tremolo.cc#...
lily/stem-tremolo.cc:182: ph = ph - ph.center ();
On 2019/04/03 08:16:22, Keith wrote:
> There is the obvious problem that line 182 destroys the effort of line 181.  
It
> reports every stem-tremolo to be centered on the central staff-line, which
makes
> line-spacing less accurate but could easily go un-noticed.

This is likely the famous "somebody must have been something" phenomenon.  Since
this does not concern the segfaulting, I am leaving this for someone else to
fix.  Ok, I'll add a TODO.