Re: [rock-devel] Building packages with make -jX

ROCK Mailing List Archives

Entire message
+ (text/plain)
Author: Sebastian Knapp
To: ROCK Development Mailing List
Subject: Re: [rock-devel] Building packages with make -jX
Am Donnerstag, 16. April 2009 18:04:39 schrieb Clifford Wolf:
> Hi giftnuss,
> On Mon, Apr 13, 2009 at 05:26:09PM +0200, Sebastian Knapp wrote:
> > for decreased build time I tried to use make with the -j switch to
> > utilize all CPU cores of my new build host. Unfortunately not all all
> > packages support this option. The list below is created simply by trial
> > and error during a build which sets $MAKE in pkg-header.
> >
> > My question is what is or would be the recommended way to deal with
> > this issue in ROCK. In my next build I plan to automate the task of
> > determining compatibilty with a customized pkgloop_action.
> > Are there plans to work around this in ROCK Linux core scripts? This
> > would be nice.
> well we could do something like the following in build_this_package():
> if [ "$makeopt" ]
> then
> eval echo "Running: $MAKE ${makeopt//\"/\\\"}"
> if [ $makej -gt 1 ]; then
> if ! eval "$MAKE -j$makej $makeopt"; then
> # FIXME: somehow mark this pkg as not -j
> compatible in cache file eval "$MAKE $makeopt"
> fi
> else
> # FIXME: somehow mark this pkg as not -j compatible
> in cache file eval "$MAKE $makeopt"
> fi
> fi
> $makej could default to '0', can be set to a different value using the
> build configuration and may be overwritten by the package, when the package
> is known to be problematic. especially it would be possible to use the
> cache files to mark the packages as problematic and use this information to
> disable make -j for those packages.
> I'd vote against a simple static list of packages that do not support -j
> since I have seen already in other projects that some packages turn out to
> have a problem only once in 100 compile cycles. So a singe try-and-error
> cycle won't be enough to show the problematic
> even when building with make -j you won't get such a good distributed load
> as with the cluster build. however - I'm not sure if the cluster build is
> working atm and combining both methods will sure give even better results.
> yours,
> - clifford

yeah, thanks for your reply. my hope was more to use flags or something
similar, but I Ä will wait to close a thread, THREAD or (type it LARGE)
thread, until stefan replies hopefully.

THX again and see you in L-E. in November?
rock-devel mailing list