On Tue, Oct 5, 2010 at 11:25 PM, Phil Holmes <address@hidden> wrote:
The other issue is - all these test cases with multiple repeats are
artificial - the real question is - does it do long _real_ scores?
I always had the feeling (though nothing to document it) that LilyPond
is much better at compiling real-world scores, no matter how huge,
than {\repeat unfold 10000} stressing tests. I've had no problem (on
GNU/Linux, that is) building either my or Nicolas' scores. Perhaps the
memory handling/garbage collection is better with real-world scores?
(e.g. with barchecks, music variables, different contexts being used,
etc?)
Similarly, here's what Han-Wen once told me when I tried compiling a
huge score imported with midi2ly:
http://lists.gnu.org/archive/html/lilypond-user/2006-10/msg00339.html
" A dirty midi file won't have note-ends aligned with barlines, which
means that the entire score will end up in one huge line without
breaks. This will undoubtedly stretch lilypond performance in
unexpected ways."
Cheers,
Valentin