Bug 151135 – [XvfbDriver] Fail to run all layout tests when X server started with -displayfd option

• Component: Tools / Tests
• Status: RESOLVED FIXED

[Carlos Garcia Campos]

Since my last debian update my xserver no longer uses the :<display-id> option, but -displayfd. The XvfbDriver uses the x server command line to check the displayas that are currently in use. This doesn't work when x server was satarted with -displayfd. This option is used to let the server find the display id available and it's written to the given file descriptor. With this option xord doesn't need to create the lock files in tmp either. The -displayfd option is also available in XVfb, so we could use it too when available. That would simplify the code, fixing also race conditions between we check for available displays and XVfb opens the connection, we wouldn't need to wait for 4 seconds after launching XVfb, and all lock files we are using won't be needed either.

[Carlos Garcia Campos]

cgarcia   1245  1.6  2.5 1118548 416480 tty2   Sl+  08:08   7:44 /usr/lib/xorg/Xorg vt2 -displayfd 3 -auth /run/user/1001/gdm/Xauthority -nolisten tcp -background none -noreset -keeptty -verbose 3
carlos   29750  0.0  0.2 479492 48224 tty3     Sl+  11:27   0:02 /usr/lib/xorg/Xorg vt3 -displayfd 3 -auth /run/user/1000/gdm/Xauthority -nolisten tcp -background none -noreset -keeptty -verbose 3

this is how Xorg is run in my laptop

[Carlos Garcia Campos]

Created attachment 265284 [details]
Patch

[WebKit Commit Bot]

Attachment 265284 [details] did not pass style-queue:


ERROR: Tools/Scripts/webkitpy/port/xvfbdriver.py:82:  multiple imports on one line  [pep8/E401] [5]
Total errors found: 1 in 2 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Carlos Garcia Campos]

Created attachment 265377 [details]
Fix coding style

[Carlos Garcia Campos]

ping reviewers, it's a bit annoying having to keep this patch around locally to be able to run the tests.

[Zan Dobersek]

Is running Xvfb via displayfd required or is that now supported because it can be run that way?

[Carlos Garcia Campos]

(In reply to comment #6)
> Is running Xvfb via displayfd required or is that now supported because it
> can be run that way?

It's not required, but there's no easy way to get a list of display ids currently used when xorg itself is run with displayfd. However, using displayfd we don't need that list at all and it fixes other issues of the current model.

[Zan Dobersek]

Would it be possible to just switch to only using displayfd? How recent is this flag?

Having two ways to run this isn't optimal. BUt if you add some test cases in xvfbdriver_unittest.py I guess it's not that big of a deal either.

[Carlos Garcia Campos]

(In reply to comment #8)
> Would it be possible to just switch to only using displayfd? How recent is
> this flag?
> 
> Having two ways to run this isn't optimal. BUt if you add some test cases in
> xvfbdriver_unittest.py I guess it's not that big of a deal either.

I agree, and I thought about it, but since I was not sure (and didn't have much time) I took the safest approach. I'll check when displayfd was added.

[Zan Dobersek]

Would it be possible to just switch to only using displayfd? How recent is this flag?

Having two ways to run this isn't optimal, but please at least add some test cases in xvfbdriver_unittest.py.

[Carlos Garcia Campos]

Created attachment 265579 [details]
Updated patch

The displayfd option is available in Xvfb since 3 years ago, so I've removed the old code and updated the unit tests.

[Darin Adler]

Comment on attachment 265579 [details]
Updated patch

rs=me

[Carlos Garcia Campos]

(In reply to comment #12)
> Comment on attachment 265579 [details]
> Updated patch
> 
> rs=me

Thanks Darin.

[Carlos Garcia Campos]

Committed r192568: <http://trac.webkit.org/changeset/192568>

[Carlos Garcia Campos]

(In reply to comment #11)
> Created attachment 265579 [details]
> Updated patch
> 
> The displayfd option is available in Xvfb since 3 years ago, so I've removed
> the old code and updated the unit tests.

I had forgotten that we build our own xserver, so we are not using Xvfb from the distro but from the jhbuild env. And the version we build 1.12.0 (released in March 2012) doesn't have the displayfd option. This shouldn't be a problem for EFL since they don't build xserver. So, I think it's better to just update xserver to a newer version or stop building our own.

[Csaba Osztrogonác]

(In reply to comment #14)
> Committed r192568: <http://trac.webkit.org/changeset/192568>

I have no idea how and why, but it broke the 
webkitpy test script on the Apple Mac bots.

[Carlos Garcia Campos]

(In reply to comment #16)
> (In reply to comment #14)
> > Committed r192568: <http://trac.webkit.org/changeset/192568>
> 
> I have no idea how and why, but it broke the 
> webkitpy test script on the Apple Mac bots.

Oh, I'll check that too.

[Carlos Garcia Campos]

I can reproduce it locally, I'm investigating.

[Carlos Garcia Campos]

Python tests should be fixed in r192569.