apport hook AssertionError in __setitem__ when trying to assign None value

Brian saw similar error reporting for compiz:
https://pastebin.canonical.com/42676/

However running unity_support_test directly doesn't cause a crash. Not sure exactly what's going on, maybe we should add a try/catch around that for now.

This works fine:

In [9]: report['UnitySupportTest'] = command_output(['/usr/lib/nux/unity_support_test', '-p'])

In [10]: report['UnitySupportTest']
Out[10]: '\x1b[0;38;48mNux: ---------------------------\x1b[0;38;48m\n\x1b[0;38;48mNux: Device pci id detection ...\x1b[0;38;48m\n\x1b[0;38;48mNux: ---------------------------\x1b[0;38;48m\n\x1b[1;32;48mNux: lspci: 01:00.0 VGA compatible controller [0300]: ATI Technologies Inc RV730 PRO [Radeon HD 4650] [1002:9498]\n\x1b[0;38;48m\n\x1b[1;32;48mNux: PCI [vendor:device]: [1002:9498]\x1b[0;38;48m\n\x1b[1;32;48mNux: Device Id: 0x1002\x1b[0;38;48m\n\x1b[1;32;48mNux: Vendor Id: 0x9498\x1b[0;38;48m\n\x1b[0;38;48mNux: ------------------------\x1b[0;38;48m\n\x1b[0;38;48mNux: Graphics device info ...\x1b[0;38;48m\n\x1b[0;38;48mNux: ------------------------\x1b[0;38;48m\n\x1b[0;38;48mNux: ------------------------\x1b[0;38;48m\n\x1b[0;38;48mNux: OpenGL version check ...\x1b[0;38;48m\n\x1b[0;38;48mNux: ------------------------\x1b[0;38;48m\n\x1b[1;32;48mNux: System OpenGL: 2.1 [Minimum Required 1.4]\x1b[0;38;48m\n\x1b[0;38;48mNux: -------------------_------\x1b[0;38;48m\n\x1b[0;38;48mNux: OpenGL extension check ...\x1b[0;38;48m\n\x1b[0;38;48mNux: --------------------------\x1b[0;38;48m\n\x1b[1;32;48mNux: Frame Buffer Object support: YES\x1b[0;38;48m\n\x1b[1;32;48mNux: ARB Vertex Program: YES\x1b[0;38;48m\n\x1b[1;32;48mNux: ARB Fragment Program: YES\x1b[0;38;48m\n\x1b[1;32;48mNux: Non Power Of Two Textures: YES\x1b[0;38;48m\n\x1b[1;32;48mNux: Rectangle Texture: YES\x1b[0;38;48m\n\x1b[1;32;48mNux: Vertex Buffer Object: YES\x1b[0;38;48m\n\x1b[1;32;48mNux: Pixel Buffer Object: YES\x1b[0;38;48m\n\x1b[0;38;48mNux: ---------------------------\x1b[0;38;48m\n\x1b[0;38;48mNux: Unity support assesment ...\x1b[0;38;48m\n\x1b[0;38;48mNux: ---------------------------\x1b[0;38;48m\n\x1b[1;32;48mNux: The system can run Unity.\x1b[0;38;48m'

so it is probably something in the __setitem__ funciton of problem_report.py of apport. It'd also be good if unity_support_test had a switch made it not use colors.

Looking at the output of unity_support_test perhaps only the last line is relevant?

On Wed, Feb 02, 2011 at 11:36:49PM -0000, Brian Murray wrote:
> Looking at the output of unity_support_test perhaps only the last line
> is relevant?

I assume in cases where the last line doesn't indicate that unity is
supported, they'd like to see the full details.

But I agree the color code stuff is ugly and should go.

If/when I ever get a few spare days I want to try to clean up the
integration of this script into the apport hook; it seems to have a
variety of problems.

As an interim solution this works around the issue and adds the data:

            ust = command_output_quiet(['/usr/lib/nux/unity_support_test',
                '-p'])
            ust = ust.replace('\x1b','').replace('[0;38;48m','').replace('[1;32;48m','')
            report['UnitySupportTest'] = ust

This issue may be worth putting in for alpha-2

I've pushed the fix (along with a try/except block to be extra sure) to git.
Checking with Kate about possible inclusion in alpha-2.

The consequence of this bug is it makes it harder for people to report bugs against compiz (and maybe xorg). Since the purpose of the alpha-2 release is specifically for testing, particularly of Unity, this seems worth it to me to include.

This bug was fixed in the package xorg - 1:7.6~3ubuntu3

---------------
xorg (1:7.6~3ubuntu3) natty; urgency=low

  * apport/source_xorg.py:
    - Tag bugs with compiz version
    - Trim out colored output from unity_support_test, which causes apport
      to choke when reporting compiz bugs.
      (LP: #710630)
 -- Bryce Harrington <email address hidden> Wed, 02 Feb 2011 16:03:57 -0800

I just got to see the (kind of) duplicate bug 716143. This assertion

    and (hasattr(v[0], 'isalnum') or hasattr(v[0], 'read'))))

would hit if the value that gets passed is neither a string nor a file object. In particular, it would error out like this if you try to pass "None".

This can happen if the command you are calling produces an error:

def command_output_quiet(command_list):
    '''
    On errors, quell error message and just return empty string
    '''
    log = command_output(command_list)
    if log[:5] == "Error":
        return None
    return log

[...]

                ust = command_output_quiet([
                    '/usr/lib/nux/unity_support_test', '-p'])
                #ust = ust.replace('\x1b','').replace('[0;38;48m','').replace('[1;32;48m','')
                report['UnitySupportTest'] = ust

I. e. if the command produces any error, ust is None, and the assignment to the report dictionary fails. Some invocations of command_output_quiet are guarded with an if, some others call programs which really Should Not Fail (tm), so these are probably okay.

I think the best approach here is to use command_output(), as we really want to know about the actual error if unity_support_test fails.

This bug was fixed in the package xorg - 1:7.6~3ubuntu9

---------------
xorg (1:7.6~3ubuntu9) natty; urgency=low

  * apport/source_xorg.py:
    - Collect jockey info in the case of proprietary drivers
    - Collect update-alternatives info in case of proprietary drivers
    - Generate nvidia-bug-report.log.gz if -nvidia loaded
    - Don't quell errors calling unity_support_test (LP: #710630)
    - Check for null when checking for upgrades. Can fail if apt.log is
      empty or unreadable. (LP: #723061)
 -- Bryce Harrington <email address hidden> Tue, 01 Mar 2011 16:58:31 -0800