[RFC] One ioslave to rule them all...

Kévin Ottens ervin at ipsquad.net
Sat Jun 25 23:46:29 BST 2005


Hello list,

I'd need some opinions about the work I've done until now and about my current 
plans about system:/. The basic idea behind this ioslave was to have one root 
allowing desktop users to easily browse the relevant places in the system. I 
want to push the idea further hence why I need to know if there are 
suggestions or objections.

In order to have this possible I developed mainly three ioslaves:
- media:/ which allows to deal with your partitions, removable media, cdroms, 
etc.
- remote:/ which allows to have "remote folders" and is possible to use thanks 
to knetattach.
- home:/ which displays home directories of the users being in the same groups 
than the current user (because it's generally more relevant to share files 
with them), and the current user home directory.

Currently system:/ points to media:/, trash:/, remote:/, settings:/ and home:/ 
(which was committed yesterday). home:/ being the last part of my great evil 
plan : hiding the real file hierarchy!

I'm now almost able to do it, in fact if I could even do it now. Some of you 
may wonder "why hiding the unixian file hierarchy??? and push use of 
system:/???". 
The answer is simply that for a desktop user this unix file hierarchy is an 
implementation detail.

So, the current status of what I've done is exactly this : if you start 
browsing using system:/ you can stay in this virtual hierarchy and do all 
your daily work using it (as a desktop user, not a sysadmin).

So the user could use only this, but some entry point to the unix filesystem 
still remain, in particular the shortcuts to the home directory...

Where I'd like to go is the following : replace $HOME with home:/$USER 
everywhere, this way the user will be "on the right track" by default (in the 
system:/ hierarchy).

Of course this would raise some issues on interoperability mainly because 
there's no consensus about the available VFS protocols between desktops. I 
currently see two problems in this area (I'll use media:/ as example since 
it's older and more known, but everything I'll present is true with home:/ as 
well) :

1) Opening a file in the system:/ (media:/) hierarchy.
When opening a file with an application, the application is launched thanks to 
its desktop file, and %u is replaced with the file URL (a media:/ URL for 
example). It'll work flawlessly with most KDE applications since they support 
KIO. But, some of them don't really support KIO completely (Kaffeine for 
example because it uses xine which only support some protocols like http and 
file). Moreover, non-KDE applications know nothing about media:/ URLs!

Then two things have been introduced:
- KIO::NetAccess::mostLocalURL() which allows KDE applications to resolve an 
URL to a file:/ URL if needed (and if possible).
- X-KDE-Protocols key in desktop files, which allows to restrict the set of 
supported protocols for an applications. Anything not present in this set is 
automatically resolved to file:/ URLs if possible before launching the 
application.

This way, we keep the compatibility with most applications, and KDE 
applications are able to have more control on the process. I then consider 
this issue as solved, the real solution would be of course to ensure that any 
non-KDE application could deal with any KDE URL but that's definitely not 
feasible currently, it would require a common VFS across all desktops, 
something that we won't have before a long time IMO.

2) Drag and Dropping a file from the system:/ (media:/) hierarchy.
It's the same kind of issue than opening a file. The main difference being 
that the application is already running, so the only counter-measure that can 
apply is KIO::NetAccess::mostLocalURL(), then only KDE applications can 
resolve the URLs to file:/ URLs... I've find no real way to make it work for 
non-KDE applications. Suggestions are welcome!


Now I'm pondering on what to do :
a) Replacing $HOME with home:/$USER right now?
b) Replacing $HOME with home:/$USER for KDE 4 only?
c) Give up the whole idea?
d) Something else?

Of course I tend to think I should apply a), but I would consider b) 
acceptable whereas I really dislike c). Of course, I'm open to any 
interesting "d)" proposal.

Any opinions? advices?

Regards.
-- 
Kévin 'ervin' Ottens, http://ervin.ipsquad.net
"Ni le maître sans disciple, Ni le disciple sans maître,
Ne font reculer l'ignorance."




More information about the kde-core-devel mailing list