215112 - "./run-mozilla.sh viewer" crashes in GTK2 build because gtk was not initialised

Status: RESOLVED INCOMPLETE

(Core Graveyard :: Embedding: GTK Widget, defect)

Description

[Mark Phillips]

User-Agent:       Mozilla/5.0 (X11; U; SunOS sun4u; en-US; rv:1.4) Gecko/20030701
Build Identifier: Mozilla/5.0 (X11; U; SunOS sun4u; en-US; rv:1.4) Gecko/20030804

The build is a gcc 3.3, non-CTL, GTK2 build.
Running "run-mozilla.sh", with no arguments, attempts to start the "viewer", but
it crashes because GTK2 was not initialised.

This bug appears to have been fixed for GTK1
(http://bonsai.mozilla.org/cvsview2.cgi?diff_mode=context&whitespace_mode=show&root=/cvsroot&subdir=mozilla/webshell/tests/viewer/unix/gtk&command=DIFF_FRAMESET&file=nsGtkMain.cpp&rev2=1.27&rev1=1.26),

but needs fixing for GTK2. 

The fix is trivial, just apply the following patch:-

--- webshell/tests/viewer/unix/gtk2/nsGtkMain.cpp.orig	2003-08-05
10:34:33.285327000 +0100
+++ webshell/tests/viewer/unix/gtk2/nsGtkMain.cpp	2003-08-05 11:04:04.562907000
+0100
@@ -35,6 +35,8 @@
  *
  * ***** END LICENSE BLOCK ***** */
 
+#include <gtk/gtk.h>
+
 #include "nsViewerApp.h"
 #include "nsBrowserWindow.h"
 #include "nsGtkMenu.h"
@@ -178,6 +180,11 @@
   signal(SIGABRT, ah_crap_handler);
 #endif // CRAWL_STACK_ON_SIGSEGV
 
+#ifdef MOZ_WIDGET_GTK
+  gtk_set_locale();
+#endif
+  gtk_init(&argc, &argv);
+
   // Initialize XPCOM
   nsresult rv = NS_InitXPCOM2(nsnull, nsnull, nsnull);
   NS_ASSERTION(NS_SUCCEEDED(rv), "NS_InitXPCOM failed");


Reproducible: Always

Steps to Reproduce:
1. cd dist/bin
2. "./run-mozilla.sh" or "./run-mozilla.sh viewer"
3.

Actual Results:  
Crash with glib2's gtype.cc complaining g_type_init had not been called.

Expected Results:  
Not crashed:-)

about:buildconfig

Build platform
target
sparc-sun-solaris2.8

Build tools
Compiler 	Version 	Compiler flags
gcc 	gcc version 3.3 	-pedantic -Wno-long-long -pthreads -pipe
c++ 	gcc version 3.3 	-fno-rtti -fno-exceptions -pedantic -Wno-long-long
-fshort-wchar -pthreads -pipe -I/usr/openwin/include

Configure arguments
--verbose --prefix=/vol/fsf.sol/mozilla-1.4-gcc
--x-includes=/usr/openwin/include --x-libraries=/usr/openwin/lib
--with-default-mozilla-five-home=/vol/fsf.sol/mozilla-1.4-gcc --with-pthreads
--with-system-jpeg --with-system-zlib --with-system-png --with-system-mng
--enable-default-toolkit=gtk2 --enable-optimize --enable-freetype2
--enable-crypto --disable-postscript --enable-tests --enable-calendar
--enable-extensions=all --enable-svg

Comment 1

[Mark Phillips]

For the curious, the actual error log is as follows:-

./run-mozilla.sh
Type Manifest File: /work/src/mozilla-gtk2-no-ctl/dist/bin/components/xpti.dat
+++ JavaScript debugging hooks installed.
nsNativeComponentLoader: autoregistering begins.
nsNativeComponentLoader: autoregistering succeeded
nNCL: registering deferred (0)

(process:27268): GLib-GObject-CRITICAL **: gtype.c:1871: initialization
assertion failed, use g_type_init() prior to this function

(process:27268): Gdk-CRITICAL **: file gdkrgb.c: line 3521
(gdk_screen_get_rgb_visual): assertion `GDK_IS_SCREEN (screen)' failed
Segmentation Fault - core dumped


Oh no!  ./viewer just dumped a core file.

Comment 2

[Mark Phillips]

Change Hardware and OS, because I can reproduce this on both Solaris and Linux

Comment 3

[Andrew Schultz]

==> gtk

Comment 4

[Christopher Blizzard (:blizzard)]

Yikes, does anyone use viewer?  I've never tried to get it working with gtk2
because I've had the embedding harness.

Comment 5

[Mark Phillips]

Actually I didn't intend to use it, just forgot to specify mozilla-bin after
run-mozilla.sh:-)

Still it should not crash.

On the other hand maybe mozilla-bin is a better default for the run-mozilla.sh
script?!

Comment 6

[Mats Palmgren (inactive)]

> Yikes, does anyone use viewer?

Everyone uses it when running the layout regression tests, don't they?  ;-)

Comment 7

[Mats Palmgren (inactive)]

Attachment: Mark's patch

Comment 8

[Christopher Blizzard (:blizzard)]

Comment on attachment 142554 [details] [diff] [review]
Mark's patch

I'm going to deny this because I don't think we're going to be maintaining
viewer at all going forward.  In fact, I think we're actively trying to remove
it at this point.

Comment 9

[Mats Palmgren (inactive)]

How am I supposed to run the layout regression tests then?

Also, I find it invaluable to be able to dump the frame tree etc. from the
menus in Viewer - are there any plans for a replacement for Viewer or will
these debug features be rolled into [a debug build of] Mozilla?

Comment 10

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Build with "--enable-extensions=default,layout-debug", and run "./mozilla
-layoutdebug"