keyboard-configuration.postinst died with SIGPIPE

I tried to reproduce this: I selected English language and French keyboard at the gfxboot menu, as you did, launched the installer as you said, left the locale in English, and kept the French keyboard layout. The installation completed successfully.

That said, regardless of my ability to reproduce this, we may be able to get somewhere by inspection. The point where it falls over is:

debconf (developer): <-- GET keyboard-configuration/variant
debconf (developer): --> 1 France - Alternative
Exception caught in process_line:
Traceback (most recent call last):
  File "/usr/lib/ubiquity/ubiquity/filteredcommand.py", line 132, in process_line
    return self.dbfilter.process_line()
  File "/usr/lib/ubiquity/ubiquity/debconffilter.py", line 281, in process_line
    if not input_widgets[0].run(priority, question):
  File "/usr/lib/ubiquity/plugins/ubi-console-setup.py", line 418, in run
    self.ui.set_keyboard_variant(self.db.get(question))
  File "/usr/lib/ubiquity/ubiquity/plugin.py", line 44, in wrapper
    return target(self, *args, **kwargs)
  File "/usr/lib/ubiquity/plugins/ubi-console-setup.py", line 307, in set_keyboard_variant
    self.keyboardDisplay.setVariant(var)
  File "/usr/lib/ubiquity/ubiquity/frontend/kde_components/Keyboard.py", line 68, in setVariant
    self.loadCodes()
  File "/usr/lib/ubiquity/ubiquity/frontend/kde_components/Keyboard.py", line 233, in loadCodes
    cfile = pipe.communicate()[0]
  File "/usr/lib/python2.7/subprocess.py", line 732, in communicate
    stdout = self.stdout.read()
IOError: [Errno 4] Interrupted system call
Mar 29 16:38:01 ubiquity: ['log-output', '-t', 'ubiquity', '--pass-stdout', '/usr/lib/ubiquity/console-setup/keyboard-configuration.postinst', 'configure'] exited with code 141

What did you do when the keyboard page was displayed? Did you change the keyboard layout or variant? I wonder if this happens if they're changed twice in quick succession, or something like that.

http://bugs.python.org/issue9867 is related to this. Although some parts of subprocess were fixed to handle EINTR in http://bugs.python.org/issue1068268, communicate wasn't.

It's extremely tempting to either monkey-patch subprocess.Popen.communicate to work around this, or to add a helper function.

Ubiquity actually doesn't install any signal handlers of its own, so EINTR had to be arising from a signal handler installed by a library. I noticed that Qt's QProcessManager constructor was installing a SIGCHLD handler without SA_RESTART. I've changed ubiquity to force that to be SA_RESTARTable, and hopefully that will stop this from happening again.

Note that SA_RESTART may well be safer anyway, as for example stdio calls can't safely be retried on EINTR.

This bug was fixed in the package ubiquity - 2.5.33

---------------
ubiquity (2.5.33) natty; urgency=low

  [ Colin Watson ]
  * Shut down the debconf-communicator instance started in DM.run. This
    should reduce the number of spurious crash reports we get.
  * KDE frontend:
    - Force the SIGCHLD handler installed by QProcessManager to use
      SA_RESTART, in the hope of avoiding problems with Python's lack of
      EINTR-safety (LP: #745014).
    - Show ext2 and btrfs in the same colour as ext3 and ext4, rather than
      in the same colour as free space.
  * localechooser/supported-locales only lists additional locales, so
    calculate the set of language packs to install using the union of it and
    debian-installer/locale (LP: #741304).
  * Do a quick check at startup to see if the debconf database is locked by
    something else, and exit slightly more gracefully than by raising an
    exception (LP: #746020).
  * GTK frontend:
    - Fix the partman plugin to convert the return value of
      gtk.TreeModel.get_value to Unicode before looking it up in a dict
      (LP: #746408).
  * Update translations from Launchpad (LP: #745924, #746500).
  * Automatic update of included source packages: flash-kernel 2.28ubuntu18.

  [ Evan Dandrea ]
  * Clarify substitution variables in the debconf templates using comments.
 -- Colin Watson <email address hidden> Sat, 02 Apr 2011 13:03:38 +0100