unity used instead of metacity when logging in as a user via ltsp

Note: By changing the login session to Classic I was able to login (Preferences > Change Session > Ubuntu Classic Desktop)

unity-support-test results:
Nux: ---------------------------
Nux: Device pci id detection ...
Nux: ---------------------------
Nux: lspci: 00:02.0 VGA compatible controller [0300]: InnoTek Systemberatung GmbH VirtualBox Graphics Adapter [80ee:beef]

Nux: PCI [vendor:device]: [80ee:beef]
Nux: Device Id: 0x80ee
Nux: Vendor Id: 0xbeef
Nux: ------------------------
Nux: Graphics device info ...
Nux: ------------------------
Nux: ------------------------
Nux: OpenGL version check ...
Nux: ------------------------
Nux: System OpenGL: 2.1 [Minimum Required 1.4]
Nux: -------------------_------
Nux: OpenGL extension check ...
Nux: --------------------------
Nux: Frame Buffer Object support: YES
Nux: ARB Vertex Program: YES
Nux: ARB Fragment Program: YES
Nux: Non Power Of Two Textures: YES
Nux: Rectangle Texture: YES
Nux: Vertex Buffer Object: YES
Nux: Pixel Buffer Object: YES
Nux: ---------------------------
Nux: Unity support assesment ...
Nux: ---------------------------
Nux: The system can run Unity.

From #ubuntu-bugs:

09:58 < stgraber> bdmurray: nope, ltsp should use a classic gnome desktop, I
                  thought didrocks fixed that
09:59 < stgraber> bdmurray: with some specific hardware it "should" work (like
                  full-intel hardware)

In the past, LTSP has simply worked like any other desktop, compiz runs its tests and if they work it starts, if the don't metacity would start instead.
That worked quite well with maverick and earlier releases and I'd hope we can get the same behavior working with Natty.

Does unity work properly in that same VM ? Is there something wrong with Virtualbox's 3D driver that would make unity fail in this case ? If so, we probably should make sure it's part of the tests done by unity/compiz to determine if the system is unity-capable.

I haven't tried running a real thin client with unity yet but I have quite a few thin clients running maverick with compiz running just fine, so I'd expect unity to work on them too (as it's a compiz plugin).

@Stéphane: so, we have a tool which is call unity_support_test located at /usr/lib/nux/unity_support_test.
This is the tool deciding if we launch unity or compiz or not…

There are some cases where this tool can crash and it's under fix. You can run manually /usr/lib/nux/unity_support_test -p to get the full output (basically, it needs):

$ /usr/lib/nux/unity_support_test -p
Nux: ---------------------------
Nux: Device pci id detection ...
Nux: ---------------------------
Nux: lspci: 01:00.0 VGA compatible controller [0300]: nVidia Corporation G71 [GeForce Go 7900 GS] [10de:0298] (rev a1)

Nux: PCI [vendor:device]: [10de:0298]
Nux: Device Id: 0x10de
Nux: Vendor Id: 0x298
Nux: ------------------------
Nux: Graphics device info ...
Nux: ------------------------
Nux: ------------------------
Nux: OpenGL version check ...
Nux: ------------------------
Nux: System OpenGL: 0.0 [Minimum Required 1.4]
Nux: -------------------_------
Nux: OpenGL extension check ...
Nux: --------------------------
Nux: Frame Buffer Object support: YES
Nux: ARB Vertex Program: YES
Nux: ARB Fragment Program: YES
Nux: Non Power Of Two Textures: YES
Nux: Rectangle Texture: YES
Nux: Vertex Buffer Object: YES
Nux: Pixel Buffer Object: YES
Nux: ---------------------------
Nux: Unity support assesment ...
Nux: ---------------------------
Nux: The system can run Unity.

So, classic GL support + FBO. Can you check that it's either working or crashing (there is a 5s timeout as well)

@Brian: did you test before changing the milestone? I think that's fixed or at least, discussed with Stéphane.

@Didier,
I tested this and verified the defect still occurs in Beta 1.

I think this is the session not being set by the ltsp client, Stéphane?

Nope. We updated LTSP in Edubuntu and Ubuntu alternate to use the same default session as the server.

So if you have Unity install on the server (default for Ubuntu desktop), it'll try using unity for the session.
This works fine with most 3D capable hardware as long as the check in compiz/unity properly detects if the hardware is indeed 3D/AIGLX capable and does the fallback to 2D gnome if it isn't.

IIRC this bug report is about the VirtualBox video card driver not working properly with unity when the X server is remote. There should ideally be a test for that which would succeed on intel hardware (where it works properly even over the network) and fail on the virtualbox video driver.

Switching back to triaged for now as it's still affecting the user.
It might be good for whoever does the detection/validation script for unity to try with remote X servers and see if there's a simple test that can be done to check whether or not compositing will work over the network.

@Stéphane, you are the one having that architecture to test. So with next nux coming tomorrow, there will be a new test tool, can you give it a try (it should handle multiple X server)?

Ok, I'll have a look (unless I can delegate to Jonathan ;)).

Ideally we should test with:
 - Intel (works fine with regular compiz when X is remote)
 - VirtualBox (doesn't seem to work according to this bug report)

And see what's the current result and what's the output of your tool.

Set as incomplete as waiting for feedback

The ltsp session is not properly falling back to 2D mode automatically with a Beta 2 install

oops. spelling correctly.... I meant NOW instead of "not"