Re: CPU_COUNT still necessary?

[John Mandereau]

Le lundi 28 décembre 2009 à 00:35 +0000, Graham Percival a écrit :
> Since lilypond-book is now fixed for parallel builds, do we still need
> an independent CPU_COUNT var?  I mean, would it be possible to use -jX
> automatically for the CPU_COUNT=x value?

According to a quick look at Make manual, it might be possible to
extract -j value from MAKEFLAGS make variable.  However, it will be
difficult to respect -j value in practice, because make and
lilypond-book don't communicate to agree on not launching more than "-j
value" jobs, where lilypond-book instance which is not waiting for
lilypond to finish and each lilypond fork count as one job; moreover,
lilypond forks require the host to have lots of RAM, whereas the role -j
option is essentially speeding up building of Documentation/examples and
the second stage (texi2html invocations), both things which consume
significantly less RAM than lilypond-book+lilypond but are well
speeded-up with -j.  That's why I'd rather keep -j and CPU_COUNT
independent.


> If not, I could manually add CPU_COUNT=4 to gub, but I'd rather avoid
> computer-specific hacks in a generalized build system.  :)

Does setting LILY_CPU_COUNT in GUB local.make still work?  If it does,
you don't have to stick any computer-specific hack.

Best,
John

signature.asc
Description: Ceci est une partie de message numériquement signée