Comment 1 Markus Keller 2015-03-25 10:58:06 EDT

If you wonder why you didn't see a test failure in the build, see bug 423561 comment 46. It's basically due to bug 210792.

Comment 2 Markus Keller 2015-03-25 11:12:01 EDT

When I try to run AllBrowserTests in my workspace, it hangs already a few lines earlier in OS.gtk_init_check(..):

OS._gtk_init_check(int[], int[]) line: not available [native method]	
OS.gtk_init_check(int[], int[]) line: 8863	
Display.createDisplay(DeviceData) line: 932

Comment 3 Leo Ufimtsev 2015-03-25 11:43:50 EDT

Is this gtk2/gtk3 specific? 

I haven't looked into this too deeply. When I run the tests (Gtk2|Gtk3) then they do seem to pass (somewhat slowly, but eventually they pass). 

Fedora 21. Gtk-3.14/Gtk-2.24

Do the tests actually fail for you?

Comment 4 Markus Keller 2015-03-26 09:18:20 EDT

(In reply to Leo Ufimtsev from comment #3)
> Is this gtk2/gtk3 specific? 

No, comment 2 happens with GTK2 (2.24.10) and GTK3 (3.4.2) on Ubuntu 12.04 32-bit. Running this is enough to reproduce:

	public static void main(String[] args) {
		new org.eclipse.swt.widgets.Shell();
	}

On the Hudson test machine, the GTK version is 2.18.9.

Comment 5 Markus Keller 2015-03-26 09:38:26 EDT

Comment 4 passes locally when I check out I20150320-0030 or the commit before bug 427776 comment 46: http://git.eclipse.org/c/platform/eclipse.platform.swt.git/commit/?id=e9c202445b7800e816318f00aed3887885c67efd

After that commit, it hangs.

Comment 6 Alexander Kurtakov 2015-03-26 09:55:15 EDT

Markus, just to be sure - this problem with GTK_TYPE_FIXED happens on both GTK 2 and 3? I ask because GTK_TYPE_FIXED is not called on GTK 3 AFAICS.

Comment 7 Markus Keller 2015-03-26 10:05:31 EDT

(In reply to Alexander Kurtakov from comment #6)
> Markus, just to be sure - this problem with GTK_TYPE_FIXED happens on both
> GTK 2 and 3? I ask because GTK_TYPE_FIXED is not called on GTK 3 AFAICS.

The hang in _GTK_TYPE_FIXED only happens on the test machine, which currently only runs GTK 2.


In my workspace, the call to "OS.GTK_TYPE_WIDGET ()" in Control's new static initializer block writes the messages below to stderr. Note that the problem still happens when I call Display.getCurrent() before new Shell(), but it doesn't happen any more when I call Display.getDefault() or new Display().

=> Looks like the static initializer just runs too early.


(process:6857): GLib-GObject-CRITICAL **: /build/buildd/glib2.0-2.32.4/./gobject/gtype.c:2722: You forgot to call g_type_init()

(process:6857): GLib-CRITICAL **: g_once_init_leave: assertion `result != 0' failed

(process:6857): GLib-GObject-CRITICAL **: /build/buildd/glib2.0-2.32.4/./gobject/gtype.c:2722: You forgot to call g_type_init()

(process:6857): GLib-GObject-WARNING **: cannot add class private field to invalid type '<invalid>'

(process:6857): GLib-GObject-CRITICAL **: /build/buildd/glib2.0-2.32.4/./gobject/gtype.c:2722: You forgot to call g_type_init()

(process:6857): GLib-GObject-CRITICAL **: g_type_add_interface_static: assertion `G_TYPE_IS_INSTANTIATABLE (instance_type)' failed

(process:6857): GLib-GObject-CRITICAL **: /build/buildd/glib2.0-2.32.4/./gobject/gtype.c:2722: You forgot to call g_type_init()

(process:6857): GLib-GObject-CRITICAL **: g_type_interface_add_prerequisite: assertion `G_TYPE_IS_INTERFACE (interface_type)' failed

(process:6857): GLib-CRITICAL **: g_once_init_leave: assertion `result != 0' failed

(process:6857): GLib-GObject-CRITICAL **: g_type_add_interface_static: assertion `G_TYPE_IS_INSTANTIATABLE (instance_type)' failed

Comment 8 Marc-André Laperle 2015-03-26 16:41:38 EDT

I could reproduce this on Ubuntu 12.04. I moved the initialization code to createWidget. I don't know if this could be done earlier but it fixes the problem for me. I also made sure that the crash that this code fixes hasn't come back.

https://git.eclipse.org/r/#/c/44711/

Comment 9 Alexander Kurtakov 2015-03-30 08:39:46 EDT

(In reply to Marc-Andre Laperle from comment #8)
> I could reproduce this on Ubuntu 12.04. I moved the initialization code to
> createWidget. I don't know if this could be done earlier but it fixes the
> problem for me. I also made sure that the crash that this code fixes hasn't
> come back.
> 
> https://git.eclipse.org/r/#/c/44711/

With this patch I get "(process:5556): GLib-GObject-WARNING **: gsignal.c:1189: unable to lookup signal "enter-notify-event" of unloaded type 'GtkWidget' " when rnning AllBrowserTests on console but the tests succeed. Would you please look into fixing it?

Comment 10 Eclipse Genie 2015-03-30 09:28:44 EDT

Gerrit change https://git.eclipse.org/r/44711 was merged to [master].
Commit: http://git.eclipse.org/c/platform/eclipse.platform.swt.git/commit/?id=4a52f8150fbdabef3de8fe2008687f7e2f438913

Comment 11 Markus Keller 2015-03-30 10:02:02 EDT

(In reply to Alexander Kurtakov from comment #9)
For me, on Ubuntu 12.04, AllBrowserTests also succeeds, but without printing anything to the console (except for the "Browser#setUp():" progress messages).

Comment 12 Alexander Kurtakov 2015-03-30 11:18:43 EDT

Markus, Mark-Andre, so this happens on Fedora 22 with GTK 3.16 and GLib 2.44. The fix I think is version guarding the callback creation like the other parts of the patch, it's not used on other versions.

Comment 13 Eclipse Genie 2015-03-30 12:09:04 EDT

New Gerrit change created: https://git.eclipse.org/r/44861

Comment 14 Alexander Kurtakov 2015-03-30 12:10:42 EDT

I would welcome testing for my patch though it's supposed to be straightforward.

Comment 15 Eclipse Genie 2015-03-31 12:45:09 EDT

New Gerrit change created: https://git.eclipse.org/r/44933

Comment 16 Eclipse Genie 2015-04-02 07:53:31 EDT

Gerrit change https://git.eclipse.org/r/44933 was merged to [master].
Commit: http://git.eclipse.org/c/platform/eclipse.platform.swt.git/commit/?id=ac5028d481c4987afadfb0539c67929c92bf84a9

Comment 17 Alexander Kurtakov 2015-04-02 07:54:17 EDT

Thanks, Mark-Andre,
with your patch warning is gone for me. Pushed to master.

Comment 18 Markus Keller 2015-04-07 06:53:22 EDT

Reopening to address Marc-Andre's comment in https://git.eclipse.org/r/#/c/44933/
> you didn't want us to add the version check here? I left it out to make it easier for you to test.

And my comment in https://git.eclipse.org/r/#/c/44861/

> Looks good, but I'd also avoid some code duplication and replace the two old occurrences of
>    if (OS.GTK3 && OS.GTK_VERSION < OS.VERSION (3, 11, 9))
> with
>    if (enterNotifyEventFunc != null)

Comment 19 Arun Thondapu 2015-04-08 10:34:11 EDT

(In reply to Markus Keller from comment #18)
> Reopening to address Marc-Andre's comment in
> https://git.eclipse.org/r/#/c/44933/
> > you didn't want us to add the version check here? I left it out to make it easier for you to test.
> 
> And my comment in https://git.eclipse.org/r/#/c/44861/
> 
> > Looks good, but I'd also avoid some code duplication and replace the two old occurrences of
> >    if (OS.GTK3 && OS.GTK_VERSION < OS.VERSION (3, 11, 9))
> > with
> >    if (enterNotifyEventFunc != null)

Fixed via http://git.eclipse.org/c/platform/eclipse.platform.swt.git/commit/?id=5b490bb962a97813a4885b327f855701a501fd7e