Community
Participate
Working Groups
We currently cannot run our swtbot tests on hudson, they all fail on the first access to a widget [1]. According to Ketan they fail most probably because there is no window manager available [2]. Can we get a window manager configured in order to solve this ? Or is there any other approach to run these tests during continuous integration ? [1] http://dev.eclipse.org/mhonarc/lists/egit-dev/msg01213.html [2] http://wiki.eclipse.org/SWTBot/CI_Server
Did you enable the XVNC plugin for your build. This will start XVNC and give you a virtual display.
I think Matthias is just getting a raw X sessions (with the black and white honeycomb background) but needs an actual desktop session. Perhaps modify .xsession of the Hudson user to start gnome-session or something?
I should note that my knowledge of how the version of SUSE we're using does this is zero.
The Xvnc option is on for the egit build but the swtbot tests can't access any widgets
(In reply to comment #4) > The Xvnc option is on for the egit build but the swtbot tests can't access any > widgets Is this an issue with how we launch/configure Xvnc w.r.t a window manager, like KDE? http://archive.cert.uni-stuttgart.de/suse-security/2005/08/msg00004.html http://linuxreviews.org/howtos/xvnc/ http://www.linuxquestions.org/questions/linux-software-2/xvnc-make-it-default-to-kde-not-twm-255680/
(In reply to comment #5) > (In reply to comment #4) > > The Xvnc option is on for the egit build but the swtbot tests can't access any > > widgets > > Is this an issue with how we launch/configure Xvnc w.r.t a window manager, like > KDE? > > http://archive.cert.uni-stuttgart.de/suse-security/2005/08/msg00004.html > http://linuxreviews.org/howtos/xvnc/ > http://www.linuxquestions.org/questions/linux-software-2/xvnc-make-it-default-to-kde-not-twm-255680/ Maybe [1] helps to set this up ? [1] http://www.linuxquestions.org/questions/linux-software-2/how-to-set-window-manager-used-for-xvnc-445967/
Ping ... How can I help to get this fixed ?
I believe this problem is also preventing some unit tests to run. I have tests which require an SWT Display, that fail with "gtk_init_check() failed" on both build and build2.
The default window manger is fvwm2, but the issue looks like it isn't being started. If I manually start a VNC process I just get a 'blank' screen, it I launch fvwm2 by hand then I see the desktop. -M.
May be completely unrelated, but see also bug 322108 for another oddity that implies "display stuff" isn't running or configured quite right. The "res" memory associated with Xvfb keeps growing. I'm hoping once this bug is fixed maybe Xvfb will be back to normal.
Is this still an issue? -M.
Yeah, I just retried and it still doesn't work (https://hudson.eclipse.org/hudson/job/egit/879/console) all ui tests fail due to this problem. Would it be an option to start window manager as a build step in our build job ? What is the most light-weight window manager on the build servers and how could I start this up ?
(In reply to comment #12) > Would it be an option to start window manager as a build step in our build job > ? What is the most light-weight window manager on the build servers and how > could I start this up ? On build.eclipse.org we start up metacity: metacity --replace --sm-disable PW
(In reply to comment #13) > (In reply to comment #12) > > Would it be an option to start window manager as a build step in our build job > > ? What is the most light-weight window manager on the build servers and how > > could I start this up ? > > On build.eclipse.org we start up metacity: > metacity --replace --sm-disable tried to run that from a shell build step before starting the maven build step running the ui tests, xvnc hudson plugin is enabled: [egit] $ /bin/sh -xe /tmp/hudson2941183789652812820.sh + echo DISPLAY::10 DISPLAY::10 + metacity --replace --sm-disable Xlib: extension "RANDR" missing on display ":10.0". it complains about missing "RANDR" and then the build job hangs forever
> it complains about missing "RANDR" and then the build job hangs forever I remember David Williams mentioning something about RANDR (R and R) and concluded it was nothing to be concerned with. If you do a search for ""RANDR" missing on display" in the comments of Eclipse Foundation/Community you'll get some more info.
(In reply to comment #14) > + metacity --replace --sm-disable > Xlib: extension "RANDR" missing on display ":10.0". > > it complains about missing "RANDR" and then the build job hangs forever The RANDR message hasn't effected any of our tests (that I could tell :-) But if not part of your vnc xstartup, you'll have to put it in the background: metacity --replace --sm-disable & In my .vnc/xstartup it looks like: exec metacity --replace --sm-disable PW
(In reply to comment #16) > (In reply to comment #14) > > + metacity --replace --sm-disable > > Xlib: extension "RANDR" missing on display ":10.0". > > > > it complains about missing "RANDR" and then the build job hangs forever > > The RANDR message hasn't effected any of our tests (that I could tell :-) > > But if not part of your vnc xstartup, you'll have to put it in the background: > metacity --replace --sm-disable & yeah, great, this did the trick, first time ever our UI tests succeeded on hudson :-) But Hudson doesn't like putting the process into background since this will leak file descriptors: [egit] $ /bin/sh -xe /tmp/hudson464268155744592334.sh + metacity --replace --sm-disable Xlib: extension "RANDR" missing on display ":11.0". Process leaked file descriptors. See http://hudson.gotdns.com/wiki/display/HUDSON/Spawning+processes+from+build for more information So we probably need to follow the other approach you proposed below. Could some hudson admin please configure .vnc/xtartup as described here: > In my .vnc/xstartup it looks like: > exec metacity --replace --sm-disable When the UI tests are started I see tons of window manager warnings: Running org.eclipse.egit.ui.test.AllTests log4j:WARN No appenders could be found for logger (org.eclipse.swtbot.swt.finder.matchers.AbstractMatcher). log4j:WARN Please initialize the log4j system properly. Window manager warning: Buggy client sent a _NET_ACTIVE_WINDOW message with a timestamp of 0 for 0x800747 (Show View ) Window manager warning: meta_window_activate called by a pager with a 0 timestamp; the pager needs to be fixed. Window manager warning: Buggy client sent a _NET_ACTIVE_WINDOW message with a timestamp of 0 for 0x800039 (Resource -) Window manager warning: meta_window_activate called by a pager with a 0 timestamp; the pager needs to be fixed. ... many more like these ... complete log is here https://hudson.eclipse.org/hudson/job/egit/891/console is there a way to get rid of these warnings ?
Webmaster could you ... please configure .vnc/xtartup as described here: > In my .vnc/xstartup it looks like: > exec metacity --replace --sm-disable then we could avoid leaking file descriptors -- Matthias
I've updated the hudson xstartup file. -M.
This doesn't work so swtbot can't create widgets this way. Hence I switched back to start metacity from my hudson job.
FWIW I have a local Hudson setup here that runs SWTBot tests and doesn't require anything special with the xvnc options. Perhaps the system's default xinitrc is starting a full GNOME session and that's what takes care of it? Maybe that's too resource intensive for Eclipse's Hudson slaves ... I can investigate our setup more if you'd like.
closing this bug as current solution to start metacity from build job is good enough for us
