Re: Issue 2356 in lilypond: Lilypond segfaults

[lilypond]

Comment #34 on issue 2356 by address@hidden: Lilypond segfaults
http://code.google.com/p/lilypond/issues/detail?id=2356

Sorry for the suggestion about axis_group_ clearing.  Now that I've read it  
closely, I completely understand what I wrote.

The reason I don't clear out the map is cuz it's actually quite small: an  
axis_group_ corresponds to a staff, lyrics, dynamics, etc..  It needs to be  
there 100% of the time, so it's correct to not be cleared out until  
finalize (it should be cleared out in finalize, however).

What this engraver does is associates axis groups to their contexts, gets  
notified when a span bar is created, and creates SpanBarStub in all  
contexts that a span bar traverses (skipping over holes if allow-span-bar  
is set to false).  Contexts need to be associated to vertical-axis-groups  
so that vertical sorting can happen (otherwise we don't know which contexts  
are higher or lower than other ones).

It seems that all that needs to be done is using some sorta SCM hash  
instead of the map, using derived_mark on this hash.

With respect to get_root_vertical_alignment, it can be called on an  
arbitrary element because the root vertical alignment of all of them will  
be the same save during the first timestep, where nothing has a root  
vertical alignment yet.

I hope this answers the "what" and the "how" (the latter of which you  
didn't ask for but I figured it wasimportant to state).  To sum up:

"you never clear out the map."
the map should not be cleared out until a finalize method.

"You take an _arbitrary_ element of the map (whatever happens to have the  
first hash bucket), then call get_root_vertical_alignment on it and work on  
the elements."
This is because the root vertical alignment can come from any of these  
elements.