The balsa/armhf tests are failing with the hirsute-proposed update

[Sebastien Bacher]

The update is failing the balsa tests for some reason.

The failure can be reproduced on ubuntu/ubuntu-hirsute-daily-arm64-server-20201125-disk1.img small instance on canonistack

$ autopkgtest-build-lxd images:ubuntu/hirsute/armhf
$ autopkgtest --shell-fail --apt-pocket=proposed=src:webkit2gtk --apt-upgrade balsa -- lxd autopkgtest/ubuntu/hirsute/armhf

starting the tests manually in the shell opened after failure it success though...

Getting the balsa source and editing debian/tests/screenshot to add some debug output

xvfb-run bash -c 'NO_AT_BRIDGE=1 balsa --check-mail & sleep 15 ; xwininfo -tree -root; xwd -root | gm convert - /tmp/debug.png; xwd -name balsa | gm convert - /tmp/lower.png; xwd -debug -name Balsa | gm convert - $workdir/current.png; killall balsa; wait %1; true'

$ autopkgtest --shell-fail --apt-pocket=proposed=src:webkit2gtk --apt-upgrade . --no-built-binaries -- lxd autopkgtest/ubuntu/hirsute/armhf

gives

'working directory: /tmp/autopkgtest.Uzr6TF/screenshot-artifacts

xwininfo: Window id: 0x50e (the root window) (has no name)

  Root window id: 0x50e (the root window) (has no name)
  Parent window id: 0x0 (none)
     1 child:
     0x200001 "balsa": ("balsa" "Balsa") 10x10+10+10 +10+10

X Error of failed request: BadMatch (invalid parameter attributes)
  Major opcode of failed request: 73 (X_GetImage)
  Serial number of failed request: 22
  Current serial number in output stream: 22
gm convert: Improper image header (/tmp/gmea3yq0).
xwd: error: No window with name Balsa exists!
gm convert: Improper image header (/tmp/gmH5TBcq).
gm compare: Request did not return an image.
autopkgtest [09:50:24]: test screenshot: -----------------------]
autopkgtest [09:50:25]: test screenshot: - - - - - - - - - - results - - - - - - - - - -
screenshot FAIL non-zero exit status 1
'

doing a ssh -X to the instance it seems to start the right dialog...

[Sebastien Bacher (seb128)]

Seems it has to do with portals? The recent depends adding has been reverted and now the test is successful?

[Alberto Garcia (berto)]

Hi, I can reproduce this problem on amd64 with WebKitGTK 2.30.6 and 2.32.0.

The problem seems to happen when xdg-desktop-portal-gtk is installed, webkitgtk 2.32 is the first version that installs the portal by default (see https://bugs.webkit.org/show_bug.cgi?id=213148 for the reasons why).

[Alberto Garcia (berto)]

I also tried running the test with WEBKIT_FORCE_SANDBOX=0 but it doesn't seem to make a difference.

[Alberto Garcia (berto)]

With the portal installed:

$ xvfb-run bash -c 'NO_AT_BRIDGE=1 balsa --check-mail & sleep 2 ; xwd -name Balsa'
xwd: error: No window with name Balsa exists!
$ xvfb-run bash -c 'NO_AT_BRIDGE=1 balsa --check-mail & sleep 2 ; xwininfo -root -tree'

xwininfo: Window id: 0x50e (the root window) (has no name)

  Root window id: 0x50e (the root window) (has no name)
  Parent window id: 0x0 (none)
     1 child:
     0x200001 "balsa": ("balsa" "Balsa") 10x10+10+10 +10+10

    -------------------------------------------------------------------

Without the portal installed:

$ xvfb-run bash -c 'NO_AT_BRIDGE=1 balsa --check-mail & sleep 2 ; xwininfo -root -tree'

xwininfo: Window id: 0x50e (the root window) (has no name)

  Root window id: 0x50e (the root window) (has no name)
  Parent window id: 0x0 (none)
     3 children:
     0x200007 "Balsa": ("balsa" "Balsa") 640x480+0+0 +0+0
        1 child:
        0x200008 (has no name): () 1x1+-1+-1 +-1+-1
     0x200003 (has no name): () 1x1+-1+-1 +-1+-1
     0x200001 "balsa": ("balsa" "Balsa") 10x10+10+10 +10+10

[Alberto Garcia (berto)]

Ok, so running Balsa without xvfb-run shows that with the portal installed the Balsa window takes way longer to appear, I'm pretty sure that's why xwd does not find the window after 2 seconds.

[Alberto Garcia (berto)]

It's g_dbus_proxy_new_sync() (triggered by g_application_register()) that times out.

Running the test with DBUS_SESSION_BUS_ADDRESS unset solves the problem for me.