On Wed, Feb 09, 2011 at 06:52:45PM +0000, Stuart Henderson wrote:
> Ports are primarily for building packages, and we want those builds to
> be repeatable rather than change depending on what you've got installed.
>
> If you want the easier user experience, use packages where possible.
>
>
> On 2011/02/09 16:19, Pascal Stumpf wrote:
> > Hi ports@,
> >
> > found the following in devel/jdk/java.port.mk:
> >
> > [...]
> >
> > .elif ${MODJAVA_VER} == "1.3+" || ${MODJAVA_VER} == "1.4+"
> > ONLY_FOR_ARCHS?= i386 amd64
> > . if ${NO_BUILD:L} != "yes"
> > JAVA_HOME= ${LOCALBASE}/jdk-1.5.0
> > BUILD_DEPENDS+= jdk->=1.5.0,<1.6:devel/jdk/1.5
> > . endif
> >
> > [...]
> >
> > So, I’m curious: Why would MODJAVA_VER = 1.[34]+ imply ‘jdk 1.5 only’?
> > This behaviour is essentially the same for 1.[56]+ too, setting
> > RUN_DEPENDS as expected but forcing a specific jdk version for
> > BUILD_DEPENDS. Can someone please explain if/why this is intended?
> >
> > IMO, for MODJAVA_VER = x.y+, it should detect whether a jdk is already
> > present on the system, and use that as both BUILD_DEPENDS and
> > RUN_DEPENDS (and if not, prompt the user to install the jdk-1.7 package).
> >
> > Cheers,
> > Pascal
> >
>
>
I’m not sure what you mean by saying the ‘build’ should be repeatable,
but I do agree that the resulting package should not depend on what was
installed at build time. So to correct myself, I think RUN_DEPENDS
should stay as-is, but the build system should not impose any artificial
restriction on which jdk version is required. Changing BUILD_DEPENDS has
no effect on the package, right?
I do know that packages are preferred, but *if* someone decides to build
from ports, he should not be forced to install 2-3 different jdk
versions (possibly with restrictive licenses) where jdk-1.7 would do
perfectly fine.
