revbumps [was Re: suggestion: Host fly-off between pkgin and nih and subsequent official integration]

[Thomas Klausner <wiz%NetBSD.org@localhost>]

On Tue, Apr 16, 2013 at 05:57:09PM +0200, John Marino wrote:
> >revdumps is completely different problem.
> 
> I wouldn't say it's unrelated.  When revbump signal an incompatible
> API changed, they are necessary.  When they are used to force
> rebuilds in pbulk, that's pro-builder, anti-user.  That's a clear
> policy decision by pkgsrc.

I disagree -- it is intended to be pro-user, but since we do not
provide (enough) binary packages, you only see the downside.

The point of the revbumps is that, if you had an abundance of binary
packages for your platform, you as a user could install a binary
package that suits the packages you already have installed.

For example, say png upgrades from 1.5 to 1.6 and its shlib major
changes, and the repository has binary packages from before AND after.
First, without the revbump, this wouldn't even be possible, since
foo-1.0 would EITHER be for png-1.5 OR for png-1.6, but due to the
revbump, foo-1.0 is clearly for png-1.5, while foo-1.0nb1 is for
png-1.6. So if we had lots of binary packages and smart tools,
"pkgtool install foo" would give you a foo package that would work
independent of if you already had png-1.6 or still had png-1.5.

I see that we're not there yet, but I don't think the method is
incorrect, just the lack of binary packages (and perhaps the tools are
not good enough).
 Thomas