Re: Issue 943 in lilypond: input/regression/slur-broken-trend.ly broken

[lilypond]

Comment #28 on issue 943 by pnorcks: input/regression/slur-broken-trend.ly  
broken
http://code.google.com/p/lilypond/issues/detail?id=943

The basic idea is as follows:

Before Joe's commit (4cabd2), the control points of the slur (set in the
calc_control_points function) were determined in a logical sequence; that  
is, the
control points of the pre-line-break slur were calculated, and the  
post-line-break
slur was notified of this direction, and then the post-line-break slur  
control points
were calculated.

After that commit, some special handling was implemented to move various  
grobs to the
top of a system via an 'after-line-breaking callback.  The method of  
determining the
"topmost" staff in a system requires calculating the staff width, and thus  
requires
calculating the control points of certain slurs.  When this occurs, it has  
the effect
of changing the order of operations, and slur directions are forwarded too  
late in
the process.

With my fix, extra precaution is taken to make sure the post-line-break  
slur has the
same direction of the pre-line-break slur.  If the order of operations ever  
changes
again, this regression will never recur (unless the internal semantics of  
Spanner
change).