[IDEA] Handling several (e.g. in a lab) machines with dpkg at once.

[Nicolás Lichtmaier <nick@debian.org>]

> In no way does that solve anything.  The reason for doing this and
> parsing the executables is to find hardcoded strings that point off to
> directories.  That is, replace the string "/usr/local/var/myname/" with
> a user-specified installation target that is different from the default
> and that is shared among many computers.

 This isn't the right answer to this problem. And this problem needs to be
addressed. I have some ideas to do it:

 PROPOSAL TO HANDLE SEVERAL MACHINES WITH DPKG AT ONCE:
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 One machine is defined as a "package master". This is the only machine in
the lab that has dpkg installed as we know. Dpkg would be modified to store
in a special place the original versions of all files that are considered
machine specific. i.e.: all files matching /var/* and /etc/* will be
installed into their dirs and into
/var/lib/dpkg/machine-specific-data/{var,etc}.

 Setting up a new client machine would just be:

     1) Mounting /usr
     2) Applying all the stored modifications (I mean installing the files
        from server's /var/lib/dpkg/machine-specific-data/{var,etc}).
     3) Storing checksums of all those files copied in 2) in a local
        place somewhere.

 When a package is upgraded in the "package master" it would send a
broadcast message to all the machines. Then it would be the time of handling
the installation of conffiles: With the locally stored checksum, the new
version, and the current local version, the common dpkg handling of
conffiles can easily be done. This could be done from the master machine,
and some intelligence could be applied here to avoid repeating the same
editions to all the machines.

 I hope somebody has read this...

 Bye!