continued: Common-VFS proposal

[Sean Middleditch]

On Sun, 2005-01-23 at 16:55 +0000, Thomas Leonard wrote:
> Sorry if the above is obvious, but it seems people are trying to replace
> existing good code with new good code, when the reason why people aren't
> using either old or new code is because they actually wanted an interface.
> We can force users to install a (small) interface, but they can choose
> whether they want a particular implementation. Same goes for
> configuration, BTW.

ROX is, then, pretty lenient.  Sorry for putting words in your project's
mouth.  ^^;  Many other projects are not so lenient, and even worse, 

There's also the problem of actually providing an interface that works
for a library that is designed for a very specific framework, and the
ugliness of trying to get two libraries not designed to work together
actually doing so.

My goal is of course to just specify an interface.  If someone wants to
write an implementation that just rides on KIOSlaves, and then write a
bridge to let KIOSlaves use gnome-vfs, that's something they'd fully be
allowed to do, assuming they can get both libraries to actually work
using a common interface.

I do fear that in this case the multiple implementations wouldn't be
helpful, however, and may in fact be harmful.  The interfaces need to
behave identically from the user perspective, or you end up with
usability problems.

I strongly believe that a new implementation design that is both
politically and technically acceptable for everyone, with a clean
interface both for individual applications using the VFS and desktop
integration, is the correct solution.

The desktop integration point is especially important here.  What I'm
thinking is to, again, not have the apps themselves do any
authentication (unless they specifically want to supply credentials) but
instead having another D-BUS service that listens for requests from the
VFS daemon for authentication requirements.  So, when KDE starts up, it
would register a service name like org.FreeDesktop.VFS.Auth.Interface.
When the VFS needed credentials to connect to a remote share (say, a DAV
share) the VFS daemon would ask that service to display the auth box,
and then get the reply.  Makes the desktop keep a single authentication
gateway (big security advantages when locked down with a secured X
extension yet-to-exist, and SELinux or something similar).