In the event that a GSettings object is being destroyed just as a change
signal is being delivered, the destroying thread will race with the
dconf worker thread for acquiring the lock on the GSettingsBackend.
If the signalling thread gets there first then the destroying thread
will block on the lock. The signalling thread adds a reference to the
GSettings object that is being destroyed and releases the lock. The
idea is that this should prevent the GSettings object from being
destroyed and thus maintain its entry in the list. Unfortunately, the
weak reference notify function is already running and as soon as we
release the lock, the list entry is removed.
The signalling thread crashes.
This bug is indicative of a serious problem encountered in many
situations where GObject instances are touched from multiple threads.
Ideally, we will move to a place where g_object_ref() is not called at
all on the GSettings object from the dconf worker thread and instead, a
dispatch will be done without holding a reference (similar to how
GAppInfoMonitor presently works). This would also prevent the
unfortunate case of someone dropping what they assume to be the last
reference on a GSettings object, only to have an already-pending signal
delivered once they return to the mainloop, crashing their program.
Making this change for GSettings (with multiple instances per thread,
the possibility of multiple backends and each instance being interested
in different events) is going to be extremely non-trivial, so it's not a
change that makes sense at this point in the cycle.
For now, we can do a relatively small and isolated tweak so that we
never access the list except under a lock. We still perform the bad
pattern of acquiring a ref in a foreign thread which means that we still
risk delivering a signal to a GSettings object that the user has assumed
is dead (unless they explicitly disconnect their signal handler). This
is a problem that we already had, however.
https://bugzilla.gnome.org/show_bug.cgi?id=710367
Change the order of the arguments on the (internal) keys_changed callback in
GSettingsListenerVTable.
This means that all functions in the table now fit the following signature:
void (* f) (GObject *target,
GSettingsBackend *backend,
const gchar *name_or_path,
gpointer origin_tag,
const gchar * const *names);
allowing the possibility of arguments ignored at the end.
This allows us to simplify our dispatch-to-thread code in GSettingsBackend,
making it a bit less generic.
So far, this should be a straight refactor.
https://bugzilla.gnome.org/show_bug.cgi?id=710367
In case an object is already monitored, we lock then return without
unlocking it which can then result in a self deadlock. So properly
unlock before returning.
https://bugzilla.gnome.org/show_bug.cgi?id=721074
On x64 builds under both Visual C++ and MinGW-64, the correct symbol for
the RunDLL call is g_win32_run_session_bus, so this will silence the
annoying RunDLL error for x64 MinGW builds as well, and will allow this
call to be executed more than once.
Thanks to Tarnyko for pointing this out.
Add missing (allow-none) annotations to
g_desktop_app_info_launch_uris_as_manager(). Tested with
python -c "from gi.repository import GLib, Gio; Gio.DesktopAppInfo.new('gcalctool.desktop').launch_uris_as_manager([], None, GLib.SpawnFlags.SEARCH_PATH|GLib.SpawnFlags.STDOUT_TO_DEV_NULL, None, None, None, None)"
which is necessary to do a launch_uris() without leaking stdout.
https://launchpad.net/bugs/1259721
D-Bus versions < 1.6.18 (i.e. all current versions) have a bug with the
path_namespace='/' match rule key. It should conceptually match everything,
but actually matches nothing. This results in no property change (or other)
signals being forwarded to the D-Bus client.
The work-around implemented in GDBusObjectManagerClient is to remove the
path_namespace match key if its value is ‘/’.
For the upstream D-Bus bug, see:
https://bugs.freedesktop.org/show_bug.cgi?id=70799https://bugzilla.gnome.org/show_bug.cgi?id=710726
It's not difficult to do; not all backends implement it, and for some
it may be difficult to implement query_info_on_read(), so let's just
do both.
https://bugzilla.gnome.org/show_bug.cgi?id=706254
Don't return children with invalid schemas from
g_settings_list_children() (ie: missing schemas or mismatched paths).
This prevents gsettings list-recursively from crashing when broken
schemas are installed on the system.
https://bugzilla.gnome.org/show_bug.cgi?id=705688
g_cancellable_disconnect will wait until any pending "cancelled"
handlers finish. This is useful because disconnecting a handler can have the
side-effect of freeing data that the cancelled handler may rely on.
Unfortunately, the code used to enforce this synchronization between
"cancelled" handlers and g_cancellable_disconnect will also cause
deadlock if the cancelled handler itself calls g_cancellable_disconect.
Obviously, if g_cancellable_disconnect is explicitly called by a "cancelled"
handler, then the "cancelled" handler is shouldering the responsibility
of not using any data that may be freed by disconnection.
Also, g_cancellable_disconnect can be called in unexpected places by
lower layers in the code (for instance as a result of g_source_destroy).
In practice, this means it's easy for deadlocks to inadvertently crop
up when using "cancelled" handlers.
For these reasons, it would be good to fix the deadlock.
This commit prevents the deadlock by allowing foregoing synchronization,
if a pending "cancelled" handler is in the same thread as the
g_cancellabale_disconnnect call.
https://bugzilla.gnome.org/show_bug.cgi?id=705395
In the real_..._async wrapper for GFile.measure_disk_usage, skip the
wrapping of the progress callback in the case that the user gave a NULL
callback to the async function. This is a performance improvement
because the sync version won't have to do continuous sampling of the
clock to issue a call to the wrapper which will then do nothing.
Unfortunately, I made this simplifying assumption when writing the
wrapper, but forgot to actually implement it when making the sync call.
As a result, the wrapper is still called, and invokes the NULL callback,
causing a segfault.
Make sure we pass NULL if the user's callback was NULL.
https://bugzilla.gnome.org/show_bug.cgi?id=707787
Make use of __wgetmainargs() on Windows so that we can get wide char
versions of the argv's that are passed in when this test program is being
invoked. This is necessary as one might enter non-ASCII, such as
CJK characters filenames and/or directories to run the test program
against, so that we can process the name(s) and pass the proper
UTF-8-encoded name(s) of the files/directories that is being tested.
https://bugzilla.gnome.org/show_bug.cgi?id=707787
It turns out that although dirent is available on mingw32 (where the
code was originally tested), it is not usable from MSVC.
Avoid portability problems by just using GDir.
Also, be careful about ensuring that we utf8-format filenames in our
error messages, and leave out the "file://" component since the strings
we're displaying are not URIs (and we don't want to make them URIs since
the extra escaping would reduce legibility).
Thanks to Chun-wei Fan <fanchunwei@src.gnome.org> for portions of this
patch and for reviews.
https://bugzilla.gnome.org/show_bug.cgi?id=707787
Virtaal installs a mime package for various .po-like file formats, one
of which has the extension .txt. This causes GLib to report ".txt"
files still as "text/plain" but no longer with complete certainty.
The result is that asserting !uncertain during the testsuite causes the
test to fail if Virtaal happens to be installed.
Remove this assertion.
On Windows and OS X, FIONREAD on a UDP socket gets the total number of
bytes available, not the number of bytes available in the next packet,
which is the more useful number (and how the function always behaved
on Linux).
On OS X, fix this by using SO_NREAD. On Windows, fix this by doing a
MSG_PEEK recv() into a giant buffer, since there is apparently no
other way to get the information.
https://bugzilla.gnome.org/show_bug.cgi?id=686786
The requested_conditions list access is not threadsafe. When passing
the socket ownership from a GSource callback to another thread, which
also creates a GSocketSource for the socket, it can happen that the
original GSocketSource is finalized at the same time as the new one
is created. This would cause inconsistencies in the requested_conditions
list and can cause assertions or completely undefined behaviour.
https://bugzilla.gnome.org/show_bug.cgi?id=705027
The RunDLL command call during get_session_address_dbus_launch() was
expecting _g_win32_run_session_bus@16 and g_win32_run_session_bus
on Win32 and Win64 respectively at least when GLib is compiled with MSVC,
not g_win32_run_session_bus@16, which caused annoying RunDLL error dialogue
boxes to show up during the use of GtkApplication (such as when running
gtk3-demo-application on Windows), prevented GtkApplication items from
being run for more than one time during the lifespan of the program,
and this also interfered with some GTK+ tests, causing them to fail.
Update accordingly to address the issue.
GNetworkAddress was allowing IPv6 scope ids in g_network_address_new()
/ g_network_address_parse(), but not in g_network_address_parse_uri().
Fix that.
Part of https://bugzilla.gnome.org/show_bug.cgi?id=669724
Convert {glib,gobject,gio}/tests to use the automake TAP driver
and test harness instead of gtester. To do so, we add a glib-tap.mk
that provides the same interface as glib.mk, except for the
reporting and coverage testing functionality. Eventually, we may
want to replace glib.mk with it. I've not yet converted the
toplevel tests/ directory, since it mixes gtestutils tests with
other binaries.
https://bugzilla.gnome.org/show_bug.cgi?id=692125
