[packagekit] Semantic issues with GetDeps (and other interface functions)

Richard Hughes hughsient at gmail.com
Tue Sep 4 10:21:50 PDT 2007


On 04/09/07, Travis Willard <travis at archlinux.org> wrote:
> On Tue, 4 Sep 2007 11:01:25 -0400, Ken VanDine wrote
> > mysql doesn't tell you that on install, but when the service is
> > started the first time.  I really think this sort of info should be
> > left to runtime, not included in the install/update.
>
> I don't disagree with you, but we're dealing with package management here.  We
> control install/update messages, however we don't control upstream software.

Yes you can, you have patches in the source format to patch the source
at build time. Red Hat replaces loads of config files this way.

> We can trivially add an install message warning the user of any number of
> things they need to be aware of, but it may be entirely non-trivial to add
> this message into the application if the original developers didn't do it in
> the first place - especially if, for example, it's a GUI application.

If it is a GUI application then it will be using gconf or the kde
equiv, which is really easy to set default vaues for.

For instance (the deps are made up, but the argument holds):

My girlfriend wants to install pidgin. pidgin deps on
caching-nameserver that deps on squid. Squid is a system level
package, and my girfriend is asked what port to run the squid proxy
instance. She runs and cries. It has to "just work". If it's not
secure in the default config, then patch the config defaults, and let
the power users use the GUI tools, webmin or just vim to do the
powerful stuff.

> Something to give users non-intrusive textual feedback during install/update
> would be very benificial.  I DON'T want or expect interaction (ie. "configure
> now? [Y/n]") once the install process has begun, but a simple feedback (ie.
> "you'll want to configure") is a boon.

What does that string mean? Seriously, if I can't understand it then
my g/f has no hope. Can it be translated into 100 languages? Can it be
qualitified in a finite number of enumerated values?

Richard.



More information about the PackageKit mailing list