Hi Shawn, just summarizing what we talked about on IRC to make sure
we're on the same page.
Shawn Walker wrote:
I have some scenarios for which I need to know what the expected
outcome is:
* two *non-preferred* publishers have the same fmri, would you expect
image-update to fail with a multiple matches error if the package has
already been installed from one of them?
* two *non-preferred* publishers have the same fmri, would you expect
an install to fail with a multiple matches error if the package has
already been installed from one of them and you did not specify the
authority?
For these two cases, my answer is the same. If the package was installed
from a specific publisher (pkg install pkg://PUBA/foo) then these cases
should work and should stick with the same publisher. If the package was
installed using the general syntax (pkg install foo), then these should
work, but switching between publishers to pick up the newest package is
probably the right thing to do.
* two *non-preferred* publishers exist on a system, one that has the
fmri in the catalog, and one that does not but the package was
installed from at some point in the past, would you expect
image-update to fail with a multiple matches error if the package has
already been installed from one of them even though both publishers
essentially own the same fmri?
I'm more ok with this breaking than with the first two cases. If the
package was installed specifically from the publisher which no longer
knows about the package, then I would expect the upgrade/install to
fail. If the package was installed without a specific publisher, then
I'd expect the install/upgrade to work seamlessly.
I think what this is all pointing to is that we need a richer way of
talking about and comparing publishers than the simple
preferred/non-preferred authority distinction. Having a simple ordering
of all publishers would resolve the first two cases. It would also give
a better set of tools for pkg to use to decide how to proceed in the
third case. There are fancier systems we could move to I'm sure, but
that seems to be a large win for little effort (of course, for all I
know an ordered list of publishers might make require rewriting 50% of
our code).
Brock
These are all related to bug 6856.
Cheers,
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
