Partial compilation (again)

[Urs Liska]

I have another suggestion, somewhat related to
http://lists.gnu.org/archive/html/lilypond-devel/2014-11/msg00139.html

I would still like to see the possibility to re-compile individual 
staves only.
And with a sloppy-line-breaking option (maybe in combination with 
system-wise output à la lilypond-book-preamble) risk is much lower that 
changes in line breaking interfere with that. (Of course that scenario 
has to be taken into account).

The benefit of such an option would be a much more responsive user 
experience, and this could be made fruitful in editors or, e.g., web 
applications: when something changes that doesn't affect line or page 
breaking LilyPond only has to recompile the relevant section (I'd say 
system).

This is an approach that is taken by Amadeus (which recompiles the 
current page), and will also be taken by the new Steinberg application. 
I don't know concretely what it will be like, but in one of the blog 
posts Daniel Spreadbury elaborated on determining the necessary minimum 
range for recompilation to find the best possible trade-off between 
realtime experience and engraving quality.

I think it would behave very similarly to the skipTypesetting approach, 
with a few differences:

- the recompiled range doesn't appear to be the beginning of the score
  (no indent, no instrumentName (but shortInstrumentName)
- barnumbers and mark numbers (and maybe other items) should also 
reflect the real values
  As they are broken items anyway there should be no impact on their 
formatting anyway

Maybe it would be good to restrict this to system-wise output because I 
don't have a clue how it should be possible to replace a system from 
within an existing PDF document. Maybe it's different with SVG output, 
though.

If sloppy line breaking is active determining the need of recompiling 
the whole document would only mean checking if the modified range still 
fits on a line.

I see two ways to approach the counter issue mentioned above:
The whole score could be parsed again, so the reduction of work load 
only affects the layout process (which is probably much more expensive 
anyway).
A LilyPond run will produce an external helper file providing meta 
information about the systems. I can imagine such information to be 
useful in other circumstances too. I'd even say with such information it 
could be possible to create such a partial compilation scheme at user-level.

I know this may be a rather big task, but I think that adding a partial 
compilation feature would be a very good thing in order to keep (or even 
make) LilyPond more competitive.

Best
Urs