ubiquity crashed with TypeError: argument of type 'NoneType' is not iterable in ubi-partman.py

Marking Triaged as I reproduced it with a recent ISO, will have a look this weekend or on Monday.

The bug is caused by atk-bridge, if it's present in GTK_MODULES, the bug happens.

Not sure where to re-assign the bug though but it seems to me like that's an atk bug and not an ubiquity bug (as ubi-partman didn't have any change related to accessibility in months).

Likely a bug in GTK's recently refactored treeview model a11y code. There may be a recent atk patch to fix it, but I need to test.

This bug is present in today's build, just tried on two machines, a Toshiba laptop and an older desktop.

I hit this today with the amd64 iso (although the bug submitter took me to #912031 initially).

If anyone wants to do a custom install, simply boot to a live session first and do it from there.

Present in 12.04 beta1 for AMD64, not possible to install.

Thanks jibel :)

linking to mis linked bug (testing):
#912031
https://bugs.launchpad.net/ubuntu/+source/partman-auto/+bug/912031
<a href="https://bugs.launchpad.net/ubuntu/+source/partman-auto/+bug/912031">link_01</a>
<a href="/ubuntu/+source/partman-auto/+bug/912031">link_02</a>

Still debugging this, as its a GTK bug. WIll be filing an upstrea bug once the bad commit is found and will link this bug to upstream bugzilla.

Is there a work-around available? I'm setting up a new system, and REALLY wanted to start with 12.04, even though it's still in Beta. I've got a partition on the disk I want to save, which is why I selected "Something else".

Yes there is a work-around. Boot the live CD, choose "Try Ubuntu", and launcht e installer from the desktop. THis bug is accessibility related, so if you follow those instructions, no accessibility will be enabled, allowing the advanced partitioner to work.

To anyone else who checks, the work-around suggested in #14 is indeed working. Thanks!

Seems like Ubuntu badly needs installer tests with Sikuli or something similar. Just run into the same problem. Looking forward to try workaround #14.

THe problem is explained in the upstream bug. It has been suggested several times that the problem be worked around in the ubiquity code itself to prevent the problem, hwoever I don't know enough of that partman code to fix it myself, and I don't have the time to learn it and fix it. Passing back to the ubiquity devs for further action.

Subscribing Martin to evaluate the pygi override from the linked bug report.

I don't necessarily mind patching ubiquity to deal with the None value but I'm guessing we aren't the only one in this situation, so a pygi workaround/fix sounds better to me.

Status changed to 'Confirmed' because the bug affects multiple users.

Reassigning back to Ubiquity. The GtkListStore documentation makes it quite clear (section "Atomic Operations") that list insertion is not atomic, and that there might be signals with None/NULL values in between. So I think for robustness this ought to be fixed in ubiquity.

That said, I'm happy to change the pygobject overrides to be atomic, as that was the recommended approach in the documentation anyway. So if you want to wait for another day or two, and I get this in soon enough, you can get away with leaving the code as it is.

Err, except that I'm unable to reassign to ubiquity -- LP says

  'None' is a binary package. This bug has been assigned to its source package 'gtk+3.0' instead.

I had a closer look at the overrides in pygobject, and it's quite some effort to rewrite them using the atomic API. I can't guarantee that I'll have them ready this week. So you could either use Gtk.ListStore.insert_with_valuesv() yourself instead of using the override API, or make the signal handlers cope with None rows.

I did the needed changes to ubiquity to use insert_with_valuesv() but I hit https://bugzilla.gnome.org/show_bug.cgi?id=671610

Are you planning on fixing that one this week or should I revert my ubiquity change so we still have a working manual partitioner when started from the live session? (it's currently in ubiquity's trunk but not release)

I meant https://bugzilla.gnome.org/show_bug.cgi?id=672065

Hello Stéphane,

Stéphane Graber [2012-03-14 20:09 -0000]:
> Are you planning on fixing that one this week

Yes, I'm working on the "empty strings in GValue arrays" bug right
now, and I'm close to a solution. Once that's fixed, I'll continue
working on the "atomic list store append" patch, which is blocked by
the former.

> should I revert my ubiquity change so we still have a working manual
> partitioner when started from the live session? (it's currently in
> ubiquity's trunk but not release)

I still recommend just getting along with None values. There will be
another signal shortly after with the actual values.

But it now seems to me that making .append() atomic (in terms of sent
signals) is realistic after all, working on it as fast as I can.

Martin

--
Martin Pitt | http://www.piware.de
Ubuntu Developer (www.ubuntu.com) | Debian Developer (www.debian.org)

For the record, I now have patches for both the "double signal" problem (this bug), as well as the underlying problem of being unable to use insert_with_valuesv(). Both are sent upstream for peer review.

This bug was fixed in the package ubiquity - 2.9.29

---------------
ubiquity (2.9.29) precise; urgency=low

  [ Oliver Grawert ]
  * disable gnome-settings-daemon power management plugin during install to
    make sure things like lid closing or accidential pressing of power or
    suspend buttons do not put us into suspend mode (LP: #948163)

  [ Stéphane Graber ]
  * ubi-partman: Ignore None cell values, we'll get another event
    when they actually contain something anyway. (LP: #939450)
  * Fix set_indicator_keymaps to always send an array of strings to
    gsettings and also go through more of the possible language codes as
    Xkl doesn't seem to always work with part2_code (LP: #955844)
  * Restrict the list of keyboard layouts added to the indicator to 4
    and try to be clever as to which one to select.
  * Only use the keyboard indicator when in greeter mode, otherwise it'd
    override any previously selected keyboard layout (from gfxboot or from
    the user session)
  * Update indicator tests to match the new code, ensuring we don't have
    more than 4 entries and that languages with variants work fine.
  * Automatic update of included source packages: console-setup
    1.70ubuntu3, partman-auto 93ubuntu22, partman-efi 24ubuntu3.
 -- Stephane Graber <email address hidden> Thu, 15 Mar 2012 18:12:41 -0400

pygobject patch committed upstream. I'll release 3.1.92 upstream today and will upload to Debian/Ubuntu afterwards.

This bug was fixed in the package pygobject - 3.1.92-1

---------------
pygobject (3.1.92-1) experimental; urgency=low

  * New upstream release.
    - Fix Gtk override API for tree models to send out only one signal after
      the data is set. (LP: #939450)
 -- Martin Pitt <email address hidden> Mon, 19 Mar 2012 17:29:07 +0100