valgrind reports an invalid read of size 1, strace shows that it traverses /usr/share/zoneinfo until the right zone (Europe/Amsterdam) is detected and crashes right after that:

stat("/usr/share/zoneinfo/posix/Europe/Stockholm", {st_mode=S_IFREG|0644, st_size=1892, ...}) = 0
stat("/usr/share/zoneinfo/posix/Europe/Amsterdam", {st_mode=S_IFREG|0644, st_size=2917, ...}) = 0
open("/usr/share/zoneinfo/posix/Europe/Amsterdam", O_RDONLY) = 20
fstat(20, {st_mode=S_IFREG|0644, st_size=2917, ...}) = 0
read(20, "TZif2\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\r\0\0\0\r\0\0\0\0"..., 2917) = 2917
close(20)                               = 0
close(19)                               = 0
close(18)                               = 0
close(17)                               = 0
--- SIGSEGV (Segmentation fault) @ 0 (0) ---

Some more gdb output:

+ Trace 194888

Line 786 is the g_ascii_isalnum() check in system_timezone_is_valid()

        for (c = tz; *c != '\0'; c++) {
                if (!(g_ascii_isalnum (*c) ||
                      *c == '/' || *c == '-' || *c == '_'))
                        return FALSE;
        }

Looks like tz is not NULL-terminated here.

In recursive_compare():

return g_strdup (file + strlen (SYSTEM_ZONEINFODIR"/"));

While reading this, I assume this won't be NULL-terminated, I see a char* and integer value concatenated and duplicated into a string.

Hmm, ignore the last comment. C is not PHP or VB, strings are not concatenated with + in C, it's a pointer operation that strips off the /usr/share/zoneinfo/ part of the timezone.

Interesting. While I try to find the bug, I have one question for ArchLinux: is there a file somewhere giving the information about the system timezone? I've read that there might be a TIMEZONE key in /etc/rc.conf?

Also interesting to see that the used file is from /usr/share/zoneinfo/posix/Europe/Stockholm. You don't have /usr/share/zoneinfo/Europe/Stockholm? I guess I should handle the posix/...
   

Archlinux sets TIMEZONE in /etc/rc.conf as "Europe/Amsterdam" as example for my own timezone setting. rc.conf is a bash script that is sourced by every /etc/rc.d/ bootscript and the rc.sysinit script that copies /usr/share/zoneinfo/$TIMEZONE to /etc/localtime (no idea why we don't symlink it though).

About the /usr/share/zoneinfo entries:
[root@jan ~]# pacman -Ql tzdata | grep Stockholm
tzdata /usr/share/zoneinfo/Europe/Stockholm
tzdata /usr/share/zoneinfo/posix/Europe/Stockholm
tzdata /usr/share/zoneinfo/right/Europe/Stockholm

The Europe/Stockholm and posix/Europe/Stockholm rules are equal, the right/Europe/Stockholm file is different.

Note that these files are also present on a debian etch system, so it's normal to have these files.

Thanks. I'll add support for /etc/rc.conf after fixing the bug.

I've committed something to strip the posix/, so we're now back to the bug.

Back to the bug: I've checked, and when using the method to compare the content of the files, it perfectly works for me, and I don't have the valgrind error...

I tested code from svn trunk (rev 11051) compiled with gcc 4.3.0 (release) with "-march=x86_64 -O2 -pipe" as CFLAGS.

using the trunk code, I still get this:

0x000000000041ff13 in system_timezone_find () at system-timezone.c:802
802	                if (!(g_ascii_isalnum (*c) ||
(gdb) print *c
Cannot access memory at address 0xff00ff
(gdb) print c
$3 = 16711935

I've been debugging a bit, removing CFLAGS from the Makefile for the clock applet, leaving only "-g", the applet doesn't crash anymore. Looks like a mis-optimization by gcc 4.3.0.

Note that you can easily compile a small test program (look at the system-timezone.c comments at the top) to debug this. If you update from trunk, please move system_timezone_read_etc_localtime_content at the top of get_system_timezone_methods since I've added support for Arch Linux.

Marking as NEEDINFO for now. Feel free to close if you're sure it's a gcc bug. Sounds weird to have a gcc bug here, though...

With -O2:

[jan@jan clock]$ gcc -O2 -g -DSYSTZ_GET_TEST -Wall -o system-timezone-get `pkg-config --cflags --libs glib-2.0 gobject-2.0 gio-2.0` system-timezone.c
system-timezone.c: In function ‘system_timezone_write_key_file’:
system-timezone.c:460: warning: passing argument 3 of ‘g_file_get_contents’ from incompatible pointer type
system-timezone.c: In function ‘files_are_identical_content’:
system-timezone.c:734: warning: passing argument 3 of ‘g_file_get_contents’ from incompatible pointer type
system-timezone.c: In function ‘system_timezone_read_etc_localtime_content’:
system-timezone.c:766: warning: passing argument 3 of ‘g_file_get_contents’ from incompatible pointer type
system-timezone.c: In function ‘system_timezone_is_zone_file_valid’:
system-timezone.c:869: warning: passing argument 4 of ‘g_io_channel_read_chars’ from incompatible pointer type
system-timezone.c: In function ‘system_timezone_set_etc_timezone’:
system-timezone.c:917: warning: passing argument 3 of ‘g_file_get_contents’ from incompatible pointer type
Starting program: /home/jan/svn/gnome-panel/trunk/src/gnome-panel-2.22.1.2/applets/clock/system-timezone-get 

Program received signal SIGSEGV, Segmentation fault.
0x0000000000401b43 in system_timezone_find () at system-timezone.c:802
802	                if (!(g_ascii_isalnum (*c) ||


Using -O:

[jan@jan clock]$ gcc -O -g -DSYSTZ_GET_TEST -Wall -o system-timezone-get `pkg-config --cflags --libs glib-2.0 gobject-2.0 gio-2.0` system-timezone.c
system-timezone.c: In function ‘system_timezone_write_key_file’:
system-timezone.c:460: warning: passing argument 3 of ‘g_file_get_contents’ from incompatible pointer type
system-timezone.c: In function ‘files_are_identical_content’:
system-timezone.c:734: warning: passing argument 3 of ‘g_file_get_contents’ from incompatible pointer type
system-timezone.c: In function ‘system_timezone_read_etc_localtime_content’:
system-timezone.c:766: warning: passing argument 3 of ‘g_file_get_contents’ from incompatible pointer type
system-timezone.c: In function ‘system_timezone_is_zone_file_valid’:
system-timezone.c:869: warning: passing argument 4 of ‘g_io_channel_read_chars’ from incompatible pointer type
system-timezone.c: In function ‘system_timezone_set_etc_timezone’:
system-timezone.c:917: warning: passing argument 3 of ‘g_file_get_contents’ from incompatible pointer type
[jan@jan clock]$ ./system-timezone-get 
Europe/Amsterdam

Net a compiler bug after all. After fixing the warnings (the parameters should not be "unsigned int" but "gsize" instead), the system-timezone-get program runs fine when compiled with gcc -O2.

On amd64, a sizeof(unsigned int) returns 4, while sizeof(gsize) returns 8. It's quite weird to see that gcc -O2 uncovers an integer overflow here (64 bit pointers don't fit in unsigned ints).

Good catch, thanks. I've committed a patch to replace all unsigned int usage with gsize. I'm closing the bug, but please test.

Man, I'm sure that I've written code elsewhere that assumes unsigned int and gsize is the same :/ If you catch any other warning of this type anywhere else in GNOME, please file bugs...

*** Bug 528442 has been marked as a duplicate of this bug. ***

*** Bug 528532 has been marked as a duplicate of this bug. ***

*** Bug 528669 has been marked as a duplicate of this bug. ***

*** Bug 528703 has been marked as a duplicate of this bug. ***

*** Bug 528865 has been marked as a duplicate of this bug. ***

*** Bug 528866 has been marked as a duplicate of this bug. ***

*** Bug 528867 has been marked as a duplicate of this bug. ***