Age | Commit message (Collapse) | Author | Files | Lines |
|
The previous attempt only updated it for 1 year, by mistake
https://bugs.freedesktop.org/show_bug.cgi?id=79548
|
|
operations
|
|
before teardown
This is to fix an assertion failure that happens in the
"/connector+ssl/econnreset/client-open" test. In this test, the server closes
the connection and the client catches up immediately and closes too.
When the client closes, it forcibly terminates the server, but in most
cases the server's socket close operation doesn't have a chance to finish
before server teardown is called, so the teardown function hits cancel
on the GCancellable that is linked with the operation and finally
the operation ends up repording an "Operation cancelled" error.
By running the GMainLoop once before calling teardown, we ensure
that the socket close operation has a chance to finish gracefully.
|
|
Since g_type_init() is deprecated in more recent glib versions, also
remove all calls of it to fix compilation. g_type_init() is not necessary
to be called anymore, it is a no-op.
https://bugs.freedesktop.org/show_bug.cgi?id=94031
Reviewed-by: Diane Trout <diane@ghic.org>
|
|
The wocky http proxy has now moved in GIO, so we no longer need
to maintain it here. I have kept the unit test for the moment,
to verify that the code in gio works the same as the one that
we used to ship in wocky.
There was one difference in the code from GIO, though. It includes
"Basic" as part of the authorization string. According to RFC this
is correct, so it looks like the wocky proxy client code was doing
this wrong. I have updated the test to reflect that.
Additionally, this commit removes support for building GIO proxy
support depending on whether GIO is recent enough or not.
We can now safely depend unconditionally on a recent enough version.
https://bugs.freedesktop.org/show_bug.cgi?id=94031
Reviewed-by: Diane Trout <diane@ghic.org>
|
|
* Add hints about what the test failure looks like
* Update the example crl update command line.
* Instead of having two copies of the CRL, replace one with a symlink.
This way when one is updated, both are updated.
https://bugs.freedesktop.org/show_bug.cgi?id=79548
Reviewed-by: George Kiagiadakis <gkiagia@tolabaki.gr>
|
|
Reviewed-by: George Kiagiadakis <gkiagia@tolabaki.gr>
|
|
It looks like gnutls has just renamed their struct types to have a _t
at the end of their name
|
|
* Replace with new call soup_session_new ();
* Indicate minimum libsoup version
https://bugs.freedesktop.org/show_bug.cgi?id=96383
Reviewed-by: George Kiagiadakis <gkiagia@tolabaki.gr>
|
|
This callback can deal with self being destroyed but it was trying to
access self->priv before early returning.
|
|
Recent (Markdown-based) gtk-doc doesn't like <code> spanning
more than one line, causing a build failure.
Reviewed-by: Guillaume Desmottes
|
|
The callback returns FALSE so the source will be invalidated.
Recent GLib now raises a critical when attempting to remove an invalid ID,
making some jingle tests failing.
(cherry picked from commit 01fbaec365cc58d6f3de46ce3f54f6413f3ec0f9)
|
|
|
|
All Automake products automatically depend on their SOURCES and
everything identifiable as a file in their LDADD, unless you set their
DEPENDENCIES, which take precedence (and have "replace", not "add",
semantics). As a result, setting DEPENDENCIES is actually harmful here:
it makes the tests and examples not depend on their own source code!
For the one test that has non-trivial extra dependencies,
use EXTRA_x_DEPENDENCIES, which has the desired semantics.
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=67953
Signed-off-by: Simon McVittie <simon.mcvittie@collabora.co.uk>
Reviewed-by: Vivek Dasmohapatra <vivek@collabora.co.uk>
|
|
The porter test asserted that if you cancelled the sending of a stanza
after it had already been (sent and) received, the send reported
success, not cancellation; and the SASL auth test asserted that if
you closed a connection at around the same time that a cancellable
had been cancelled, the close reported success, not cancellation.
However, recent GLib seems to be either more careful about deferring the
results of async operations to an idle, or more consistent about
reporting the cancellation as an error even if the operation's
success had already been recorded. As a result, these operations
reported cancellation. To avoid that, delay the cancellation a little.
[Mike Ruprecht clarified on IRC that "more consistent about
reporting the cancellation as an error" is one of the changes from GIO
being ported to GTask in 2.36. -smcv]
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=67900
Signed-off-by: Simon McVittie <simon.mcvittie@collabora.co.uk>
Reviewed-by: Vivek Dasmohapatra <vivek@collabora.co.uk>
|
|
The very next thing we did with it was to give it to GSocketConnection,
which would turn on non-blocking again; so this code wasn't doing
anything except harming our portability.
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=67900
Signed-off-by: Simon McVittie <simon.mcvittie@collabora.co.uk>
Reviewed-by: Vivek Dasmohapatra <vivek@collabora.co.uk>
|
|
GLib now follows Unicode Corrigendum 9, which clarifies that
libraries shouldn't prohibit non-characters. We were assuming it did.
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=67900
Signed-off-by: Simon McVittie <simon.mcvittie@collabora.co.uk>
Reviewed-by: Vivek Dasmohapatra <vivek@collabora.co.uk>
|
|
GSocket configures its underlying fd to be in non-blocking mode, and
implements blocking calls by select()ing (or equivalent) first.
If we break this assumption, most test cases in wocky-connector-test
hang in a recv() that should have been non-blocking.
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=67900
Signed-off-by: Simon McVittie <simon.mcvittie@collabora.co.uk>
Reviewed-by: Vivek Dasmohapatra <vivek@collabora.co.uk>
|
|
I'm not sure whether this ever worked correctly, but the build currently
fails on Debian unstable, and this fixes it.
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=67875
Signed-off-by: Simon McVittie <simon.mcvittie@collabora.co.uk>
Reviewed-by: Guillaume Desmottes <guillaume.desmottes@collabora.co.uk>
|
|
Some were already checked, but as assertions: this is
(now) public library API, so downgrade to g_return_if_fail().
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=65131
Signed-off-by: Simon McVittie <simon.mcvittie@collabora.co.uk>
Reviewed-by: Xavier Claessens <xavier.claessens@collabora.co.uk>
|
|
wocky_jingle_session_parse() advances the session's state
machine. In particular, it may cause termination, which
causes the session to be removed from the factory's
hash table, which may cause its last ref to be released.
Until recently, this would have gone unnoticed, but
wocky_jingle_session_acknowledge_iq() now emits a signal
from the session (to check whether it has the "is Google
webmail" quirk), and that causes a check that it is in fact
still a valid session object.
Also correct a misleading comment spotted while debugging
this: priv->sessions owns both key and sess.
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=65131
Signed-off-by: Simon McVittie <simon.mcvittie@collabora.co.uk>
Reviewed-by: Xavier Claessens <xavier.claessens@collabora.co.uk>
|
|
The Google webmail client currently starts calls like this:
<iq type="set" ...>
<jingle xmlns="urn:xmpp:jingle:1'"
action="session-initiate" ...>...</jingle>
<session xmlns="http://www.google.com/session"
type="initiate" ...>...</session>
</iq>
(This isn't valid XMPP Core, because 'An IQ stanza of type "get"
or "set" MUST contain exactly one child element', but we can
tolerate it.)
When called, it also echoes the contents of the sender's IQ back to the
sender. It appears that this is how, when it makes an outgoing call,
it determines which dialect the recipient wants to use: the recipient
echoes the appropriate child element.
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=65131
Reviewed-by: Xavier Claessens <xavier.claessens@collabora.co.uk>
|
|
This avoids compilation failing because the new(ish) g_thread_new
wasn't available in our target version. telepathy-gabble indirectly
depends on GLib 2.32 anyway, via telepathy-glib 0.20.
Reviewed-by: Xavier Claessens <xavier.claessens@collabora.co.uk>
|
|
|
|
It's checked elsewhere for XMPP 1.0 servers, which can either
use "old SSL" or perform STARTTLS. Legacy Jabber can only use
"old SSL", which is similar to https - connect to a separate port,
typically 5223, and start speaking SSL - so if the connection was
ever going to be encrypted, by this point it already would be.
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=65036
Reviewed-by: Sjoerd Simons <sjoerd.simons@collabora.co.uk>
|
|
|
|
|
|
These pass because 565f2ed54.
|
|
https://bugs.freedesktop.org/show_bug.cgi?id=61433
|
|
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=61792
Reviewed-by: Simon McVittie <simon.mcvittie@collabora.co.uk>
|
|
|
|
|
|
Previously there were a couple of ways this could crash:
* First, if you had a <field type='hidden' var='FORM_TYPE'> with no
<value> child, field->default_value would be NULL so
g_value_get_string (field->default_value)
would critical and return NULL and then you'll be in undefined hell.
* Having fixed that, the code to sort data forms by FORM_TYPE would also
crash because it happens before the FORM_TYPEs have been validated and
used g_value_get_string() without checking and then strcmp()ed the
possibly-NULL result.
I actually have a work-in-progress branch that makes this all a lot less
hairy by adding, among other things, a wocky_data_form_get_form_type()
function which can do the validation for us. But better to fix the
crashes before refactoring them away.
https://bugs.freedesktop.org/show_bug.cgi?id=61433
|
|
https://bugs.freedesktop.org/show_bug.cgi?id=61433
|
|
https://bugs.freedesktop.org/show_bug.cgi?id=61433
|
|
Notice that some of the paths set priv->result to NULL before completing
it, and some do not. I particularly like the one in
establish_session_recv_cb which assigns priv->result to a tmp variable
but forgets to set priv->result to NULL!
I hoped this would make the connector re-usable between successful
connection attempts, but it doesn't because priv->conn is not cleared
out so it crashes when the stream is establed the second time you call
wocky_connector_connect_async(). At least all the code paths crash
consistently now, though! (Previously if you had connected to a pre-XMPP
Jabber server, wocky_connector_connect_async() did not immediately
reject a second call and would crash later in the same way this now
does.)
|
|
This made the assertion that failed confusing: the message was about
'identity', but it was actually a test of the 'resource' property.
|
|
I changed the default value for WockyConnector:resource to NULL in
5fb069a, but didn't update the test accordingly.
|
|
|
|
Previously, there was no way to leave the resource entirely up to the
server: while the binding code deals correctly with priv->resource being
NULL, it was impossible to make it NULL.
But the random resource generation was broken, too: the RNG was never
seeded so the resource always came out the same.
|
|
|
|
The changes are:
* payment-required is removed;
* not-authorized is added to the schema (it was already defined in
Wocky);
* policy-violation is added. XEP-0086 doesn't define an error code
mapping for it, so I chose 406 "Not Acceptable".
|
|
|
|
|
|
|
|
Lots of the signals had inline comments describing their parameter
lists, but no gtk-doc comments.
|
|
Astonishingly, putting the instance struct before the class struct makes
WockyMuc's signals and properties show up in the generated
documentation.
|
|
|
|
This was renamed to wocky_porter_register_handler_from(), but its
documentation was not updated. Also, the arguments to the function were
wrong: there was a NULL where the JID should be, and the JID appeared
later in the wrong place.
While we're here, fix some formatting earlier on in the same comment. I
think using @ rather than the <!-- --> trick is more legible.
|
|
|