Re: Prefer luatex for documentation

[Jonas Hahnfeld]

(I am not sure why these were sent to lilypond-user, moving back to
lilypond-devel where they belong)

On Mon, 2022-11-21 at 23:14 +0100, Jean Abou Samra wrote:
> Le 21/11/2022 à 22:10, Jonas Hahnfeld via Discussions on LilyPond 
> development a écrit :
> > > And whether we can just *require* LuaTeX and stop looking for pdfTeX
> > > and XeTeX altogether?
> > I did a few measurements for the case of building the LilyPond
> > documentation and, in terms of speed with the "CI configuration" (no
> > extractpdfmark and using the Ghostscript API), LuaTeX seems to position
> > itself between pdfTeX, which remains the fastest, and XeTeX. So at
> > least in my opinion, this would be a viable path and we could just
> > always build with LuaTeX.
> 
> Interesting. Do you still have the precise timings around?

Yes, of course:

XeTeX
real    3m55.427s
user    33m11.744s
sys     1m45.433s

pdfTeX
real    3m44.164s
user    31m43.151s
sys     1m15.535s

LuaTeX
real    3m51.352s
user    33m2.542s
sys     1m30.938s

This is with "-j12 CPU_COUNT=12". I didn't post the absolute numbers
initially because with parallelism, the PDF compilation may actually be
hidden behind the invocations of lilypond-book (for "real" time) and
the "user" time may be influenced by how long parallel lilypond-book
processes have to wait for their turn to lock the lybook-db.

Conclusion: If this is to become a serious proposal, we should test
some configurations with less parallelism (shared runners are
sequential, my runner has two cores, and Karlin's runner (new btw since
last week) has four cores).

Jonas

signature.asc
Description: This is a digitally signed message part