[packagekit] Debconf and PackageKit Was Re: Packagekit and Ubuntu

[Colin Watson]

On Mon, Nov 23, 2009 at 05:46:41AM -0800, Daniel Nicoletti wrote:
> It's really fun how this far more complex draft was somehow
> accepted by Colin, as Richard said this is a nightmare to debug,
> and there is NO benefit to not use passthrough frontend,
> my DBus frontend made in just two weeks accomplished all these
> matters and doesn't even needed to worry about NameOwnerChanged.
> 
> Afaik Colin was on vacations, probably used that to go to UDS,
> I really would like him to answer my last email about using passthrough
> and doing a really simple change in debconf-comunicate would put an
> end to all these problems.
> 
> Really, the solution is simple I just need an OK to don't
> write code that does not get accepted

I've been intending to give you a more detailed response, but have not
had time yet. However, on inspection of the code you sent previously, I
found much the same issues as I explained to you at some length on
#debian-devel a while back. Joey may feel differently (I don't want to
put words into his mouth), but I'm afraid my previous comment stands; I
do not intend to accept any code that creates a completely new UI
framework for debconf. This approach divides the already scarce effort
apparently available for writing good-quality debconf UIs (witness the
fact that nobody's got round to writing a KDE4 UI module yet - your code
is *only* usable in a PackageKit context, not in general; and
furthermore we'd have to write substantial new code in order to use the
same approach with GNOME).

Using passthrough is in fact substantially less code than your approach;
the unmarshalling layer is trivial, at only a couple of dozen lines of
largely table-driven code, and the marshalling layer is even simpler at
a handful of lines. This is much cleaner than inventing an entirely new
set of D-Bus methods which need to be handled entirely separately, I
intend to get it all put together before demonstrating it, though, since
my impression so far is that it's going to be a lot easier to *show* how
I think it should work rather than trying to explain it.

You've repeatedly mentioned two weeks as a small amount of development
time; so far it took me two *hours* (while at a conference, surrounded
by distractions) to get something working, although I still need to
rearrange it to match what I discussed with James and others, and the
feedback from Richard. I'm a lot happier with this since it will
actually be maintainable as we evolve debconf.

-- 
Colin Watson                                       [cjwatson at debian.org]