David Brownlee <abs@formula1.com> writes:
> 
> 	As I understand it, with the buildlink changes a package should be
> 	able to build with a different version of itself already installed.

This is only partly true.  If you use xpkgwedge, then this is
absolutely true for packages that set USE_BUILDLINK_ONLY.  If not,
then this is false for any packages that USE_X11 or that depend on
packages that set USE_X11BASE.  When I finish the implementation of
USE_BUILDLINK_X11, which links X11 headers and libraries into
${BUILDLINK_DIR}, then your statement will always be true for packages
that set USE_BUILDLINK_ONLY and for packages that set
USE_BUILDLINK_ONLY, USE_X11{,BASE}, and USE_BUILDLINK_X11.

> 	If this is the case (or when this is the case :), would it make
> 	sense to change the behaviour of 'make update' to perform a 'make'
> 	before the 'pkg_delete -r'? This would reduce the time a machine
> 	is without the package, and eliminate the possibility of a build
> 	issue leaving the machine without that package (though obviously
> 	not helping for any packages that depend on the rebuilt one).

You would have to try to detect whether the appropriate cases where
this may safely be used (described above) exist, then run your
modified "make update".  It may be better to do this after buildlink
is more mature.

	Cheers,

     -- Johnny C. Lam <lamj@stat.cmu.edu>
        Department of Statistics, Carnegie Mellon University
        http://www.stat.cmu.edu/~lamj/