Re: Incremental compiling

[Joseph Rushton Wakeling]

On 09/08/12 08:35, David Kastrup wrote:
> No, it doesn't.  As opposed to TeX which makes its _vertical_ break
> decisions sequentially and independently, LilyPond uses global
> optimization.  One can still save the local decision tree, but it may
> take a number of pages before it locally collapses to a single choice.

Vaughan was suggesting what _could be done_ rather than what _is_ done.

Does Lilypond currently do _any_ separation of the work into different jobs?  I 
ask because it seems likely that anywhere there's a known page break, you can 
carry out the engraving independently.  So e.g. if I have a 3-movement work and 
it's set that each movement will start on a new page, Lilypond could in 
principle engrave each of those movements in a separate job.  That could be in a 
separate thread, or Lilypond could just do each movement sequentially -- the 
point is that optimizing the layout for each movement does not need to depend on 
the others.

Currently I could do that manually by having each movement as a separate LP 
project and building them separately, but it could be useful to have Lilypond 
detect those opportunities automatically.

Regarding global optimization, this sounds like it might become unmanageable for 
scores more than a certain size.  How does memory usage scale with work size 
(roughly)?