On Wed, 22 Jun 2005, Sean Davis wrote:

> why is it necessary to nuke
> every package before rebuilding? why not build, and IF IT BUILDS, then nuke
> and reinstall? That seems to be the saner course of action to me. That way
> I'm not left with half the packages I had before because I updated one
> relatively insignifigant package that some weird dependency chain linked to
> KDE, and then to everything else.

I don't see how this will consistently work. If you want to update a
package and it requires new headers/libraries/features from a dependency
package, then that dependency packages must be rebuilt.

And it doesn't make sense to reinstall the dependency package if not
compatible with all the packages depending on it -- so they should all be
removed.

Also, we (usually) don't want to reinstall a package when it is still
installed, so we deinstall it first. And since other packages may depend
on now missing functionality, we make sure they are deinstalled also.

I have no problem with the removal of packages first. Using pkgsrc on
production system like this would not work for me. I do use packages made
from pkgsrc on production systems.

As for me I rarely use make update and I don't use pkg_comp. I just use a
different system to do my builds. Or I use a chroot environment which I
manually build.

I also frequently use IGNORE_RECOMMENDED=yes on systems (that don't create
binary packages) and it makes it easier.

I have also been using this mk/bsd.pkg.mk addition:

 		${ECHO_MSG} "${_PKGSRC_IN}> Required installed package
$$pkg: $${found} found"; \
 	else								\
 		${ECHO_MSG} "${_PKGSRC_IN}> Required package $$pkg: NOT
found"; \
+		found=`${PKG_ADMIN} -d "${PKGREPOSITORY}" lsbest "$$pkg"`
; \
+		if ${PKG_INFO} -qe ${dep:C/:.*$//:C/[=><-].*$//:Q} ; then
\
+			${ECHO_MSG} "${_PKGSRC_IN}> Older package that is
not good enough was found already installed.";	\
+		fi;							\
+		if [ -n "$$found" -a -n "${MY_BIN_INSTALL}" ] ; then	\
+			${ECHO_MSG} "${_PKGSRC_IN}> Installing existing
$$found ... " ; \
+			${SU_CMD} "${PKG_ADD} ${_BIN_INSTALL_FLAGS}
$$found" ; \
+		else							\
+			if [ -n "$$found" ] ; then			\
+				${ECHO_MSG} "${_PKGSRC_IN}> $$found
exists, but not using." ; \
+			fi;						\
 		target=${DEPENDS_TARGET:Q};				\
 		${ECHO_MSG} "${_PKGSRC_IN}> Verifying $$target for $$dir";
\
 		if [ ! -d $$dir ]; then					\
@@ -3650,6 +3677,7 @@
 			${SETENV} _PKGSRC_DEPS=",
${PKGNAME}${_PKGSRC_DEPS}" ${MAKE} ${MAKEFLAGS} $$target
PKGNAME_REQD=\'$$pkg\' MAKEFLAGS="" || exit 1; \
 			${ECHO_MSG} "${_PKGSRC_IN}> Returning to build of
${PKGNAME}"; \
 		fi;							\
+		fi;	\
 	fi
 .      endfor	# DEPENDS


Sorry it is out of context, as I have other changes too.

But basically it is an improvement on the bin-install target, as it does
not run as root the whole time and it only rebuilds packages if needed. It
has problems, but has saved me a lot of time. (I make sure I always
generate binary packages for reuse on this system.)

> I'm tired of it. So: what'll it be? money? hardware? or should I just act
> like nothing is wrong with the update target, like the pkgsrc developers do?

I can't speak for that. But I am guessing that "make update" must work for
some of the developers.


 Jeremy C. Reed

 	  	 	 BSD News, BSD tutorials, BSD links
	  	 	 http://www.bsdnewsletter.com/