Re: Windows speed improvements - update

[Urs Liska]

Am 07.05.2015 um 13:06 schrieb David Kastrup:
> Phil Holmes <address@hidden> writes:
>
> > I've just started working again on some mensural music with associated
> > modern transcription.  The ancient score is about 8 pages long.  With
> > 2.18.2 one of the pieces takes 97 seconds to compile.  With 2.19.19 it
> > takes 37 seconds.
> >
> > Fantastic!
> A bit too fantastic.  It would be interesting what change was the major
> contributor here.
>
> If it's Windows-specific only, it's probably related to font access.

I don't think it's Windows specific as there have been reports on other 
systems too.
I just compiled

\repeat unfold 800 { c' d' e'8 d' c'4 }

three times with averages of

2.16.2: ~21.5 sec.
2.19.15: ~ 16 sec.
2.19.17: ~ 14 sec.
2.19.20: ~ 12 sec.

on Debian.

However, there's one more thing to consider here, particularly with the 
recent experiences wrt .ly compilation failures, is if there's a 
significant difference between binary releases and custom builds. From 
the above only 2.19.20 is self-compiled. But a self-compiled modified 
2.19.16 also takes only 12 sec. As do modified builds of 2.17.3 and 
2.19.6! Unfortunately my binary release of 2.18.2 is broken ATM.

So according to my (limited) tests it seems that LilyPond itself isn't 
really speeding up. What *is* reducing compilation time seems to be 
related to the binary releases only.
Whatever that means ...

Urs