Issue 3205: opening bar check causes crash if \score contains the \midi block

Issue 3205: opening bar check causes crash if \score contains the \midi block

This derives Bar_check_iterator from Music_iterator rather than from
Simple_music_iterator.

In
\score {
  { | d }
  \midi { }
}

the bar check iterator is being called while only the \Global context
exists.  That causes Score_performer::one_time_step to be called
without getting Score_performer::prepare to be called previously,
probably because the Score context is created at the wrong time.  The
Score_performer is not prepared for this situation.

I have no idea how to fix Simple_music_iterator, why it exists in the
first place and is written like it is, and why this appears to work.

[Keith, 2013-02-27 07:07:58]

I tried skipTypesetting and showFirstLength in different situations.  They seem
to work.

It would be nice to know at least /how/ this change causes 
Score_performer::prepare() to be called.

On the other hand, it seems that the bar-check-iterator should not have caused
Score_performer::one_time_step() to be called at all.

[dak, 2013-02-27 08:28:51]

On 2013/02/27 07:07:58, Keith wrote:
> I tried skipTypesetting and showFirstLength in different situations.  They
seem
> to work.
> 
> It would be nice to know at least /how/ this change causes 
> Score_performer::prepare() to be called.
> 
> On the other hand, it seems that the bar-check-iterator should not have caused
> Score_performer::one_time_step() to be called at all.

It is not the bar-check-iterator that calls one_time_step as you can see by
leaving off the note.  But the bar-check-iterator causes
Score_performer::prepare to be missed.

The code in score-performer.cc (as well as score-engraver.cc except that the
latter does not mind missing prepare all that much) registers listeners for
Prepare and OneTimeStep _when_ the Score context is created.

A mere \context Score { | d } is already enough to prevent the segfault.  But if
you start right with | without a score context, Prepare and OneTimeStep only get
registered once d turns up, and it would appear that the Prepare event has gone
by that time and is no longer seen by the listener.

I have no clue what makes the standard Music_iterator different here.  The only
other use of Simple_music_iterator is with \partial, and \partial wraps its
music event in (descend-to-context ... and thus is impervious to this problem
(the raw music event in itself will likely crash LilyPond equally well).

I have no idea what Simple_music_iterator is for.  Possibly for getting
triggered at most once per time step, but then it should likely be possible to
get this effect without needing to actually compare moments.