https://bugzilla.gnome.org/show_bug.cgi?id=690553

The results were postponed to show as long as there were new notifications
about added contacts, which could take quite long for many matched items.
This shows the results with smaller timeout and without postponing.

Directory monitoring is a nice-to-have feature, but is not supported in
some contexts.  If we fail to create a GFileMonitor, leave a breadcrumb
on the console to indicate something whent wrong, but don't return an
error status.  That would cause the whole registry process to terminate.

(cherry picked from commit af31c05e)

This was reported at Red Hat's bugzilla:
https://bugzilla.redhat.com/show_bug.cgi?id=910252

To have this fully working, users should edit the account and save
the changes (even when not done any) by clicking the OK button
in the account editor, thus the source will receive its extension
and the next start the account will be recognized and auto-updated.

Zarafa server with disabled SSL v2 rejects connection when
SSL v2 compatible hello is sent, thus disabled this by default.
After all, SSL v3 should be used in general these days anyway.

Use environment variable CAMEL_SSL_V2_HELLO=1 in case your server
requires the old behavior.

Apparently e_dbus_object_get_source_removable() does not check its
argument for NULL, and scratch sources have no internal EDBusObject.

(cherry picked from commit a8eb939a)

Apparently e_dbus_object_get_source_writable() does not check its
argument for NULL, and scratch sources have no internal EDBusObject.

(cherry picked from commit 8797493a)

When receiving the complete address book, a client has to use
e_book_client_is_self() on every EContact to find the one which is the
"self" contact. Calling e_book_client_get_self() instead does not
work, because that would create the self contact if none exists (not
desired for an app which just reads!).

The problem with e_book_client_is_self() is that it creates and
destroys a GSettings instance for the self UID each time the method is
called. In addition to reading the value over and over again, this
also triggers two D-Bus messages (AddMatch and RemoveMatch) - clearly
bad for performance.

To solve this problem, this patch caches the GSettings instance in a
static variable. It is protected by a mutex, to keep the function
thread-safe. The downside is that the instance is never going to be
freed.

It would be better to attach the GSettings instance to the EBookClient
instance, but because e_book_client_is_self() doesn't get a pointer to
that, this is not possible without an API change.

CalDAV backend depends on local cache when a client ask for an event,
which is usually fine, but when it comes to a meeting invitations,
with which the server can add event to the calendar on its own,
the invitation might not be found in the cache, thus try whether
it's available on the remote machine, before reporting Object Not Found.

GTK+ uses (G_PRIORITY_HIGH_IDLE + 20) for redrawing operations, which is
actually a slightly lower priority than G_PRIORITY_HIGH_IDLE.  Therefore
for our purpose, G_PRIORITY_HIGH_IDLE is sufficient.

(cherry picked from commit 820dcf6c)

This reverts commit b6f0db12.

The commit contains not a single comment as to why these custom priority
values are being used.  The rationale needs to be documented in the code,
either at each call point or preferably at a centralized priority value
definition.

I recently added a Gmane (NNTP) account and have only a few newsgroup
subscriptions.  Yet when I quit Evolution, Camel tries to synchronize
all 13,872 newsgroups which literally takes days to complete.

This adds a CAMEL_STORE_FOLDER_INFO_SUBSCRIBED flag to the request for
a folder info tree while synchronizing + expunging.

Note, CamelIMAPXStore will ignore the CAMEL_STORE_FOLDER_INFO_SUBSCRIBED
flag if its "use-subscriptions" setting is FALSE, which is what we want.

(cherry picked from commit 48b9d17d)

(cherry picked from commit ad1b7cd1)

In situations when a synchronous operation was cancelled, but the response
was already piled in main context the client could receive two replies, one
from the reply, the other from the cancelled operation, effectively accessing
invalid memory in the second response. This may address also other similar
situations caused by cancelled operations.

In certain situation, the backend can be freed while the callback
is still in action, then the factory can crash inside
cal_backend_http_ensure_uri() or such.

Signed-off-by: Gheyret Kenji <gheyret@gmail.com>