On Wed, 13 Jan 2010 19:37:14 1100, Joe Neeman wrote:
> It's nice to see someone else touching the page-breaking code
The more I dig into it, the more questions I have.
Back in January, we talked about vertical space estimation in the case of Prob
(i.e. markup); now I am investigating vertical space estimation for the case of
Grob (i.e. score). The problem is that when the user tries to fit more systems
on the page by decreasing the spacing/padding, the final page layout does take
it into account, but it has no effect on the estimations at the page-breaking
stage. The visual result is that the lines do get closer together, but the
number of lines per page still remains the same, leaving a big empty gap at the
bottom of the page.
This problem is caused by the calculation of ext_len in the "for (i=...)"
cycle in Page_breaking::min_page_count().
For illustration, I have attached two examples: simple-nolyrics.ly, and the
same score with added lyrics, simple-lyrics.ly. When there are no lyrics, the
estimation works as expected. The ext_len in this case is 7.947, and the
specified padding of 1.2 causes the systems to be nicely squeezed together.
Now if we add lyrics, the final spacing and the estimation start to diverge
dramatically. In the final layout of the page, the lyrics do not cause a
significant change to the distance between the systems: the lyrics just sit in
the gap which was already there. But the estimator, during the page breaking,
get confused: for each line, the rod is lengthened by ext_len padding, and
ext_len in this case is as much as 10.203. So the page breaker thinks that the
systems are much taller than they will finally prove to be when the final
spacing is done.
Trying to do something about this ext_len problem, I traced the origin of the
ext-len back to the Y-ext of the Grob (fill_line_details() calls
sys->pure_height(), which calls get_property_data("Y-extent")). Bien sûr, the
Y-extent includes the whole height of the lyrics.
I am not sure what can be done to reconcile the estimator with the actual
spacing, so that users can get tight spacing. My understanding is that the
final spacing actually tries to avoid skyline collisions; the page breaker
ignores the actual shape of the skyline and treats the systems as solid
rectangular blocks. So the only possible workarond would be to manually
indicate the ext_len to the page breaker (something like, a block variable in
the score layout) -- extremely ugly, and I would only opt for that if I knew
for sure that all other options were exhausted.
Boris
_______________________________________________
bug-lilypond mailing list
bug-lilypond@gnu.org
http://lists.gnu.org/mailman/listinfo/bug-lilypond
