Re: Summary of KDE filesystem discussion

[Jarno Elonen <elonen@iki.fi>]

Hi - again,

I already requested comments for the following filesystem layout on the KDE 
list but I guess this list is a bit more appropriate for the subject.

Now that I've thought a bit more about it, I think this possibly could work 
with ordinary symlinks, too, and moving to the new layout would not mess up
previous installations too badly.

I'm still not proposing it to be implemented to Debian (don't think I'm in 
a position to do that anyway), but would like to have some generic discussion 
on wether it could be possible to renew the filesystem layout in the future 
and especially to make it more suitable for "ordinary users" and desktop 
environments?

So, here goes the original mail:

--- clip ---

Just as a side note (NOT as a proposition by any means!):
what's really so wrong in C:\program files style? Of course, on
open systems, instead of vendor specific directories, there should be some
other subdirectory policy (lsm for example?).

If there were a way to remove symlinks when the original file is removed,
I think the following structure would be the easiest to understand and
administrate:

  + usr
    + bin
      + qtcups -> ../qtcups/bin/qtcups
      + nano -> ../nano/bin/nano
    + sbin
      + traceroute -> ../traceroute/bin/traceroute
    + qtcups
      + etc (conf)
      + share (data)
      + bin (binaries)
      + doc (man, info)
    + nano
      + etc
      + bin
      + doc
    + traceroute
      + etc
      + bin
      + doc

...or even:

    + kde -> kde2.2
    + kde2.2
      + kmail
      + konqueror
    + kde3
      + kmail
      + konqueror
    + bin
      + kmail -> ../kde/kmail/bin/kmail

...or EVEN:
    + usr
      + X11R6
        + kde -> kde2.2
        + kde2.2
      ...

Then, you could easily try different versions and remove whole
packages manually without having to guess where their may have installed
their binaries, libraries, configuration files, data, documents etc
etc. Of course, this is pure fantasy since there are no "2-way links". If
there were, however, this model could well be made reasonably backwards
compatile.

Does it make any sense? I.e. have I missed some important aspect of Unix 
here?

- Jarno