Fix two problems:
1) If g_socket_service_stop is called before the accept call is requeued,
then the reference count won't decrease and this code will hang forever:
while (G_OBJECT (service)->ref_count == ref_count)
g_main_context_iteration (NULL, TRUE);
2) Sometimes the testcase fails (maybe 1 in 200 times for me):
GLib-GIO:ERROR:socket-listener.c:73:connection_cb: assertion failed
(G_OBJECT (service)->ref_count == 2): (3 == 2)
Aborted (core dumped)
The problem is that depending on ordering, cancellation of the async
listener can require further main context iterations before it releases
the reference on the socket service. Furthermore, in some cases, it
requires at least one iteration.
https://bugzilla.gnome.org/show_bug.cgi?id=712570
Add a property to GNetworkMonitor indicating the level of network
connectivity: none/local, limited, stuck behind a portal, or full.
The default implementation just returns none or full depending on the
value of is-available.
https://bugzilla.gnome.org/show_bug.cgi?id=664562
Add G_IO_ERROR_CONNECTION_CLOSED as an alias for
G_IO_ERROR_BROKEN_PIPE, and also return it on ECONNRESET.
It doesn't really make sense to try to distinguish EPIPE and
ECONNRESET at the GLib level, since the exact choice of which error
gets returned in what conditions depends on the OS. Given that, we
ought to map the two errors to the same value, and since we're already
mapping EPIPE to G_IO_ERROR_BROKEN_PIPE, we need to map ECONNRESET to
that too. But the existing name doesn't really make sense for sockets,
so we add a new name.
https://bugzilla.gnome.org/show_bug.cgi?id=728928
This is a convenience method for creating a GNetworkAddress which is
guaranteed to return IPv4 and IPv6 loopback addresses. The program
cannot guarantee that 'localhost' will resolve to both types of
address, so programs which wish to connect to a local service over
either IPv4 or IPv6 must currently manually create an IPv4 and another
IPv6 socket, and detect which of the two are working. This new API
allows the existing GSocketConnectable machinery to be used to
automate that.
Based on a patch from Philip Withnall.
https://bugzilla.gnome.org/show_bug.cgi?id=732317
We intend to keep the list of poll records sorted by (integer) file
descriptor, but due to a typo we are actually keeping it sorted by
pointer address of the GPollFD.
Fix that.
https://bugzilla.gnome.org/show_bug.cgi?id=11059
g_settings_has_signal_handlers() checks whether any of the signals has
pending handlers. However, g_signal_has_handler_pending() matches on
exact detail, even when passing 0. Subscribing to one of GSettings'
signals with a detail will fail this check and never connect to the
backend.
Fix this by calling has_handler_pending() with the key as detail as
well.
https://bugzilla.gnome.org/show_bug.cgi?id=740848
Add a GSocketListener test program. Currently the only test is a
regression test for bug 712570 (based on a standalone bug reproducer
provided by Ross Lagerwall).
If all users of a GThreadedSocketService release their references to the
service while a connection thread is running, the thread function will
release the last reference to the service which causes the finalize to
deadlock waiting for all threads to finish (because it's called from the
thread function).
To fix this, don't wait for all threads to finish in the service's
finalize method. Since the threads hold a reference to the service,
finalize should only be called when all threads are finished running (or
have unrefed the service and are about to finish).
https://bugzilla.gnome.org/show_bug.cgi?id=712570
If SSL 3.0 has been disabled (at the host, application, or library
level), then the "use-ssl3" property becomes a "fail-immediately"
property.
Despite the name, the point of the property wasn't really specifically
to use SSL 3.0; it was to allow fallback when talking to broken
servers that do SSL/TLS negotiation incorrectly and break when they
see unexpectedly-high version numbers. So if we can't fall back to SSL
3.0, then the "use-ssl3" property should fall back to TLS 1.0 instead
(since there are hosts that will reject a TLS 1.2 handshake, but
accept a TLS 1.0 one).
glib-networking is being updated to implement that behavior, so update
the documentation here.
https://bugzilla.gnome.org/show_bug.cgi?id=738633
In commit 8ff5668, we are subscribing the GSettings backend later, but this
meant that we need to initialize cache_lock earlier, as we might try to
use that lock before a change notification is issued to subscribe the
backend, which would then cause an access violation if we are trying to
read GSettings values, as that lock is used to access the Windows Registry.
Initialize cache_lock once we initialize the GSettings Registry backend,
and delete it upon finalize, so that g_settings_read_from_backend() can
proceed normally, even if the GSettings backend is not yet subscribed.
https://bugzilla.gnome.org/show_bug.cgi?id=740413
GSettings objects begin watching for changes as soon as they are created
in order that they can emit the "changed" signal.
In the case of dconf, if we want to be able to emit the changed signal,
we need to go on the bus and add some match rules. This requires
creating the dconf helper thread and also requires initialising GDBus
(which creates another thread).
Some users of GSettings are never interested in the "changed" signal.
One of these users is the glib-networking code that gets run every time
a new network connection is created.
Some users are reporting that they are annoyed that simply establishing
a network connection would spawn two extra threads and create a D-Bus
connection.
In order to avoid doing unnecessary work for these simple uses, delay
the subscription until we know that we will actually need to do it.
We do this in a simple way, using a simple argument: in order for the
user to care that a value changed then they must have:
1) watched for a change signal; and then
2) actually read a value
If the user didn't actually read a value then they cannot possibly be
interested in if the value changed or not (since they never knew the old
value to begin with and therefore would be unable to observe that it
ever changed, since they have nothing to compare the new value with).
This really is a behaviour change, however, and it does impact at least
one user: the 'monitor' functionality of the GSettings commandline tool,
which is interested in reporting changes without ever having known the
original values. We add a workaround to the commandline tool in order
to ensure that it continues to function properly.
It's also possible to argue that it is completely valid to have read a
value and _then_ established a change signal connection under the
(correct) assumption that it would not have been possible to miss a
change signal by virtue of not having returned to the mainloop.
Although this argument is true, this pattern is extremely non-idiomatic,
and the problem is easily avoided by doing things in the usual order.
We never really talked about change notification in the overview
documentation for GSettings, so it seems like now is a good time to add
some discussion, including the new rules for when one can expect change
signals to be emitted.
https://bugzilla.gnome.org/show_bug.cgi?id=733791
This should already work according to the documentation, but doesn't
because main_options is consumed before the check in
g_application_parse_command_line().
Fix by moving the check for main_options up.
https://bugzilla.gnome.org/show_bug.cgi?id=740157
It simplifies a little bit some code that inserts data relative to a
GList location, that might be NULL for the tail of the queue. A NULL
sibling is probably less useful for insert_after(), so it's more for
consistency with insert_before().
https://bugzilla.gnome.org/show_bug.cgi?id=736620
The win32 headers do:
#define interface struct
which is just evil and breaks other code that assumes it can use
"interface" as a variable name. gnetworking.h was supposed to be doing
"#undef interface" after including the win headers, but it did it too
soon, resulting in it getting redefined by a later include. Fix this.
https://bugzilla.gnome.org/show_bug.cgi?id=738551
g_tls_certificate_new_from_file() was only loading the complete chain
if it was fully valid, but we only meant to be validating that it
formed an actual chain (since the caller may be planning to ignore
other errors).
https://bugzilla.gnome.org/show_bug.cgi?id=729739
Fix a hang due to overflow by using unsigned numbers and explicitly
checking if the number overflows to zero. This also fixes the previous
logic which assigned an int which may be negative to an unsigned number
resulting in sign extension and strange results.
Use gsize rather than int to allow for large buffers on 64 bit machines.
https://bugzilla.gnome.org/show_bug.cgi?id=727988
Fix a hang due to overflow by using unsigned numbers and explicitly
checking if the number overflows to zero. This also fixes the previous
logic which assigned an int which may be negative to an unsigned number
resulting in sign extension and strange results.
Use gsize rather than int to allow for large streams on 64 bit machines.
https://bugzilla.gnome.org/show_bug.cgi?id=727988
If a given fd is being polled by multiple sources, we used to pass it
multiple times to g_poll(), which is technically illegal (and not
supported by the select()-based fallback implementation of poll() in
gpoll.c), and also made it more likely that we'd exceed the maximum
number of pollfds.
Fix it to merge together "duplicate" GPollFDs. The easiest way to do
this involves re-sorting context->poll_records into fd order rather
than priority order. This means we now have to walk the entire pollrec
list for every g_main_context_query() and g_main_context_poll(),
rather than only walking the list up to the current max_priority.
However, this will only have a noticeable effect if you have tons of
GPollFDs, and we're already too slow in that case anyway because of
other O(n) operations that happen too often. So this shouldn't change
much (and the new poll API will eventually let us be cleverer).
Remove some win32-specific code which did the same thing (but was
O(n^2)).
https://bugzilla.gnome.org/show_bug.cgi?id=11059
There are two consistent interpretations that could be taken for memory
handling of the 'invocation' parameter passed to the method_call() virtual
function of GDBusInterfaceVTable
- A reference is passed (transfer full) to the method_call() virtual function,
and that reference is then passed (transfer full) to the return_value/error
functions on GDBusMethodInvocation.
- An internal reference is retained from the point where method_call() is called
until the return_value/error function is called.
Since the return_value/error functions were already marked (transfer full),
we use the first interpretation, annotate the invocation parameter of
method call as (transfer full) and describe this in the documentation, along
with the idea that you are always supposed to call one of the return_value/error
functions.
See bug 738122 for the leak this caused in GJS.
https://bugzilla.gnome.org/show_bug.cgi?id=738259
Clarify in the documentation that a GSource created with
g_cancellable_source_new() must be explicitly removed from its
GMainContext before the GCancellable can be finalised.
This could be a common way of leaking GCancellables.
https://bugzilla.gnome.org/show_bug.cgi?id=737259
da053e34 broke the tls-certificates test by requiring the backend to
implement g_tls_certificate_verify() (which the test TLS backend
didn't). Add a trivial implementation to make the test pass again;
we'll need something more complicated when we add tests that are
supposed to get errors.
So shortcut it.
I wrote this patch less as a performance optimization and more as a
clarification, so that people looking at the code can be assured of this
invariant.
https://bugzilla.gnome.org/show_bug.cgi?id=738374
