Failing !is_empty assertion and tremolos

[Aaron Hill]

I think I have narrowed down the source of the problem.

From stem-tremolo.cc:175-177:

>  Interval ph = stem->pure_y_extent (stem, 0, INT_MAX);
>  Stem_info si = Stem::get_stem_info (stem);
>  ph[-dir] = si.shortest_y_;

There is an implicit assumption that si.shortest_y_ will always be 
within the interval ph.  Otherwise, the final line will produce an 
inverted interval.  And later on in the function, the !is_empty 
assertion will fail when ph.center() is called.

Here is a snippet where I have \overridden Stem.stem-info with a 
procedure that detects and corrects the value that ultimately leads to 
an invalid interval:

%%%%
\version "2.19.82"
\relative a {
  \override Stem.stem-info = #(lambda (grob)
    (define msg "\nwarning: shortest_y ~a outside interval ~a")
    (let ((si (ly:stem::calc-stem-info grob))
          (ph (ly:stem::pure-height grob 0 0)))
      (if (not (<= (car ph) (cadr si) (cdr ph)))
        (format #t msg (cadr si) ph))
      (list (car si) (max (car ph) (min (cdr ph) (cadr si))))))
  a32 8..:32 b32 8..:32 c32 8..:32 d32 8..:32 e32 8..:32
  f32 8..:32 g32 8..:32 a32 8..:32 b32 8..:32 c32 8..:32
  d32 8..:32 e32 8..:32 f32 8..:32 g32 8..:32 a32 8..:32
}
%%%%

Of course, this is not to say that Stem::calc_stem_info is definitely in 
the wrong, but its output is not always what Stem_tremolo::pure_height 
would seem to expect.


-- Aaron Hill