Installer bails on Lucid due to poor interaction with bind9 and resolvconf

[jdreed]

On Lucid installs (possibly only LiveCD-based installs?), resolvconf is installed, but /etc/resolv.conf is not normally a symlink. This causes the bind9 initscript to barf when it tries to call resolvconf.

After setting a static IP with NetworkManager, my /etc/resolv.conf is not a symlink, and has a "# Generated by NetworkManager" line at the top. If I then install -workstation, when it gets to installing bind9, it tries to call resolvconf in the initscript, which barfs with: "Error: /etc/resolv.conf must be a symlink"

Killing the postinst process causes the install to finish normally.

Possible workarounds include:

• Identifying this situation (having resolvconf installed, but not used) in the Debathena install script and fixing it or giving the user the opportunity to do so.
• Setting bind9/run-resolvconf to false in debconf

Since this only affects Lucid, there's clearly an upstream bug somewhere, but I don't know where. Insight would be appreciated concerning whether or not this is resolvconf's fault, bind's fault, or networkmanager's fault.

[jdreed]

Turns out this is a bug when resolvconf and bind9 are installed at the same time (I mistakenly thought resolvconf was installed before I started the debathena-workstation installation).

So perhaps the workaround is easy -- if -workstation is selected, install resolvconf ahead of time.

[jdreed]

Fixed in r24786.

Evan notes that resolvconf was downgraded to universe, which is why we're seeing this for the first time.

[jdreed]

This needs testing and then should be pushed out (but not this week)

[amb]

This fix had been deployed.