I have supplied a means of reproducing the problem that doesn't depend on my MIDI software under development, and verified that it does indeed reproduce the problem. 

In any given recording session, it appears there is a 1-in-3 chance of hitting the problem.  I thought it might depend on the speed of the machine used, but my testing yesterday did not support that idea. 

The thing that's different about this particular piece of music, is that the chords channel uses a sustained instrument voice, so in changing from a major 7th chord to a minor 7th chord (and back), it only changes the 3rd and 7th interval notes of the chord, while the other notes continue playing. 

I still suspect that the problem is related to missing events during the recording process, but I don't have hard evidence to support that idea.  I do know that one of the cases I recorded where it didn't crash or hang, also didn't miss a single note.  But I also have cases where a few chord notes were missed that did not encounter the crash/hang at the end of recording. 

Some recording sessions where it crashed/hung actually left notes playing that should have been silenced, so note-off events are being missed as well. 

The thing that may be different about my MIDI creation, is that the note-on (and note-off) events for the individual notes of the chords are sent back-to-back over the channel, with no delay time whatsoever in-between. 

By the way, I do want to take the time to express my appreciation for your creating and maintaining Rosegarden.  It is an amazingly full-featured sequence-editor, and I really like it for those reasons.  I have been using it successfully since Ubuntu 8.04 (and even earlier than that). 

Great work, all of you! 

And Many Thanks,

- Aere

On Sun, 2012-06-24 at 06:55 -0400, D. Michael McIntyre wrote:
On Saturday, June 23, 2012, Aere Greenway wrote:

> I think that shortening the piece may cause it to not encounter the
> problem, as my short recordings (where I made a mistake, and stopped
> recording early) haven't failed yet.

This reminds me that somebody or other reported a bug a good long time ago, 
whereby Rosegarden would do something evil only while recording extremely long 
improvs that were running for like 20 minutes or something.

I never could reproduce that one.  I was thinking about using LEGO to build a 
robot to hit keys for 20 minutes or something as a laboratory style control, 
because improving for that long is just too random, and it would never reveal 
the underlying pattern.

I wouldn't be surprised if this ghost you two are chasing is related to that, 
and that problem existed back maybe six years.