On 10/09/17 14:33, Brooks Davis wrote: > On Sun, Oct 08, 2017 at 08:42:35AM -0700, Rodney W. Grimes wrote: >>> >>>> On Oct 7, 2017, at 16:33, Sean Bruno <sbr...@freebsd.org> wrote: >>>> >>>> Author: sbruno >>>> Date: Sat Oct 7 23:33:14 2017 >>>> New Revision: 324406 >>>> URL: https://svnweb.freebsd.org/changeset/base/324406 >>>> >>>> Log: >>>> Fix symlink if_igb.ko in -current such that its relative and doesn't >>>> end up with non-standard DESTDIR information in its symlink. This >>>> can happen very trivially if the release scripts are used. >>>> >>>> Sponsored by: Limelight Networks >>>> >>>> Modified: >>>> head/sys/modules/em/Makefile >>>> >>>> Modified: head/sys/modules/em/Makefile >>>> ============================================================================== >>>> --- head/sys/modules/em/Makefile Sat Oct 7 23:30:57 2017 (r324405) >>>> +++ head/sys/modules/em/Makefile Sat Oct 7 23:33:14 2017 (r324406) >>>> @@ -23,6 +23,6 @@ CFLAGS += -I${SRCTOP}/sys/dev/e1000 >>>> #CFLAGS += -DDEVICE_POLLING >>>> >>>> afterinstall: >>>> - ln -sf ${DESTDIR}${KMODDIR}/${KMOD}.ko ${DESTDIR}${KMODDIR}/if_igb.ko >>>> + cd ${DESTDIR}${KMODDIR} && ln -sf if_em.ko if_igb.ko >>> >>> While this is definitely better, the same thing could be achieved via: >>> >>> ln -sf if_em.ko ${DESTDIR}/.../if_igb.ko >> >> Should this be using SYMLINK+= as I thought using ln -sf was a breakage >> for meta mode stuff? > > At the very least it should use INSTALL_LIBSYMLINK or INSTALL_RLIBSYMLINK. > Direct use of ln(1) is always a bug in install targets and ideally we > would make it fail. (It breaks NO_ROOT installs by failing to add an > entry to the METALOG file). > > -- Brooks >
I'm more than happy to make changes here. The closest example I can see in the tree from a Makefile is some of the things in etc: ${INSTALL_SYMLINK} ../var/unbound ${DESTDIR}/etc/unbound Is that, more or less, something y'all would like to see? sean
signature.asc
Description: OpenPGP digital signature