MAKE_JOBS and openjdk6

[Greg Lewis]

On Sat, Aug 28, 2010 at 02:28:02PM +0200, David Naylor wrote:
> On Friday 20 August 2010 17:12:42 Anonymous wrote:
> > Anonymous <swell.k at gmail.com> writes:
> > > David Naylor <naylor.b.david at gmail.com> writes:
> > >>> %%
> > >>> Index: java/openjdk6/Makefile
> > >>> 
> > >>> @@ -266,3 +267,6 @@ post-install:
> > >>>  	@${CAT} ${PKGMESSAGE}
> > >>>  
> > >>>  .include <bsd.port.post.mk>
> > >>> 
> > >>> +
> > >>> +# XXX: use `?=' in bsd.port.mk
> > >>> +_MAKE_JOBS=
> > >>> %%
> > >> 
> > >> Yes, I prefer this approach.  See attached for the patch that does this.
> > >>  I will file a PR about this shortly.
> > > 
> > > I've filed ports/148754 about defining empty _MAKE_JOBS so it's not
> > > forgotten.
> > 
> > That PR was recently committed. So, you can try to resurrect ports/148753.
> 
> I've had a look at openjdk6 and it appears it really is MAKE_JOBS_UNSAFE.  
> There are portions of it that are able to use make jobs and those are compiled 
> using HOTSPOT_BUILD_JOBS.  
> 
> I suggest that either:
>  - openjdk stops using HOTSPOT_BUILD_JOBS and declares itself unsafe, or

It has already declared itself unsafe.

>  - declare itself make jobs safe and use HOTSPOT_BUILD_JOBS for those parts 
> that can use it
> 
> Attached is a patch that achieves the latter suggestion.  
> 
> The problem with the port as it stands now is that it breaks with 
> FORCE_MAKE_JOBS, does not honour MAKE_JOBS_NUMBER and that it will consume a 
> lot of resources when building, more so than what is reasonably expected.  
> Simply declaring the port make jobs unsafe does not fix the resource 
> consumption that some programs/scripts may take into account.  
> 
> Taking the first option will result in slower build times when the port is able 
> to build faster.  
> 
> Taking the second option results in overriding a 'private' variable.  There is 
> precedent in ports for using that 'private' variable.  With the recently 
> committed changes using the 'private' variable is less intrusive.  
> 
> I recommend the second option.  It allows the port to build as fast as 
> possible, to honour MAKE_JOBS_NUMBER and does not employ any hacks.  

I would argue that overriding a private variable is a hack (other ports
doing it doesn't make it not a hack).

Alternative patch attached which seems to achieve the same result from
my perspective without overriding _MAKE_JOBS.

> diff -ur /usr/ports/java/openjdk6/Makefile openjdk6/Makefile
> --- /usr/ports/java/openjdk6/Makefile	2010-07-15 22:29:26.000000000 +0200
> +++ openjdk6/Makefile	2010-07-15 22:33:45.000000000 +0200
> @@ -48,6 +48,7 @@
>  
>  # java extracts directly to the cwd
>  WRKSRC=		${WRKDIR}
> +MAKE_JOBS_SAFE=	yes
>  
>  USE_GMAKE=	yes
>  USE_MOTIF=	yes
> @@ -145,8 +146,10 @@
>  USE_DISPLAY=	yes
>  .endif
>  
> -BUILD_JOBS_NUMBER!=	${SYSCTL} -n kern.smp.cpus
> -MAKE_ENV+=	HOTSPOT_BUILD_JOBS=${BUILD_JOBS_NUMBER}
> +.if !defined(DISABLE_MAKE_JOBS)
> +MAKE_ENV+=	HOTSPOT_BUILD_JOBS=${MAKE_JOBS_NUMBER}
> +_MAKE_JOBS=
> +.endif
>  
>  COPYDIRS=	\
>  	hotspot/src/os/linux/launcher \




-- 
Greg Lewis                          Email   : glewis at eyesbeyond.com
Eyes Beyond                         Web     : http://www.eyesbeyond.com
Information Technology              FreeBSD : glewis at FreeBSD.org
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Makefile.diff
Type: text/x-diff
Size: 643 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-ports/attachments/20100828/1470d073/Makefile.bin