Lilypond and stability [Was: K460 does not compile under 2.6.0]

[Sebastiano Vigna]

On Sun, 2005-07-10 at 20:07 +0200, Erik Sandberg wrote:
> realised - you can write your own repeat unfold! I'm no Scheme expert, 
> but I managed to write this quick hack:

Scheme hacks are not portable across versions, and certainly not fixed
by convert-ly (I just removed a snippet from LSR because of that).

In my view Lilypond is to music as TeX is to text (this might be a
narrow view, of course). One of the basic issues that Knuth wanted to
solve with TeX was to have a stable system that would compile forever
his books exactly in the same way. Everybody knows that a Word document
has a lifespan of a few years, after which the input filter start to get
crazy and the document is gone (you have still the text, but the rest
has just a pale resemblance with the original). But I can still compile
my undergrad TeX notes without problems, getting the same output. Now,
music has an even larger lifespan--technical books are often replaced by
newer books, but music is never replaced.

The problem I see with regressions is that people with a large body of
Lilypond music will grow more and more scared of recompiling with new
versions. I mean--what can you do? Check measure by measure that
hundreds of pages of music are still getting out right? That your
carefully chosen page/line breaks are still reasonable?

Of course, it's a tradeoff--if you want to make the layout algorithms
better, you change the output. And clearly Lilypond is constantly
improving, which is a Good Thing(TM). I wonder, however, why several
scores on mutopia that were compiled on Lilypond 1.x are not recompiled
with newer versions. They could clearly get better, so my guess is that
the author is scared of losing control of what will happen.
-- 
Ciao,

                                        seba