automatic keyboard layout detection loops and messes up screen

cdebconf-keystep (0.7) hardy; urgency=low

  * Port fixes from cdebconf-entropy (LP: #172807):
    - Switch the handler symbol to the "cdebconf_" namespace.
    - Use the proper namespace to access newt frontend API.
    - Remove the usage of dlsym() and dlopen() now that frontend symbols are
      loaded globally.
    - Bump Depends on cdebconf accordingly.
  * Change Maintainer to <email address hidden>.
  * Add Vcs-Bzr field.
  * Fix Section.
  * Fix configure options so that it doesn't always think we're
    cross-compiling.
  * Remove unused debhelper command cruft.

 -- Colin Watson <email address hidden> Thu, 29 Nov 2007 16:32:50 +0000

According to http://archive.ubuntu.com/ubuntu/dists/hardy/main/installer-amd64/current/images/udeb.list the installer has built against 0.8, but the bug still happens on the current images.

I don't seem to be able to reproduce this on i386 now. I wonder if this is a different bug.

Could you attach /var/log/syslog from after the failure, please?

(Or a screenshot would probably do if extracting syslog is too hard.)

I still get the looping: I press 'y', 'w', and am back to the 'do you want to detect...' question. /cdrom/pool/main/c/cdebconf-keystep/* has version 0.8.

On my first boot I did not get screen corruption, the question dialog was clean. On the second boot I did get the garbled screen again (see screenshot in comment 2), so this looks like a race condition.

I attach /var/log/syslog.

I'm not quite sure what was going on back in late November / early December, but I tried it with the final http://archive.ubuntu.com/ubuntu/dists/hardy/main/installer-amd64/current/images/netboot/mini.iso and it worked fine, so I think this has gone away now. The syslog Martin posted ("undefined symbol: newt_create_window") is definitely a sign of something still using cdebconf-newt-detect-keys 0.6 for some reason.