On Saturday 27 February 2010, Kevin Kofler wrote:
If they Obsolete something else, then they're not really new
packages.
I that's the blanket generalization I read it as, I don't agree with it, but
meh.
Well, true, new packages which Provide some common virtual Provides
like
bluez-dbus-pin-helper also need the same scrutiny as upgrades to explicit
packages. That's not the common case though, [...]
Common or not, it is one occurrence of what I wanted to point out as you have
more than once in this thread made a broad claim that new packages just
can't/won't break existing setups, and still make in a slightly less broad
form below.
New packages which don't Obsolete existing packages or Provide
existing
provided names cannot cause any of the above.
Those "provided names" also include the package's Name, and all files
shipped
in the package, or more generally, anything that other packages or mechanisms
(e.g. package groups) can have a dependency on so that it gets pulled in
without explicitly asked. Whether the "provided name" is existing or not is
irrelevant, a dependency on it can spring to life any time, including at a
time when it causes the package containing the name to be installed without
being explicitly asked.
And FWIW, if you think outside of the Fedora box, that set is (perhaps not
strictly, but practically) infinite, and I suppose there is no active ongoing
effort/process to check all of these even within Fedora.
Anyway in my opinion it is not really relevant to this discussion whether new
packages may end up being installed without explicitly being asked for. I
think it's better to just treat and push them like other updates, be it
through testing or directly to stable.