| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
If a CM is initially offline when the Tpf.PersonaStore is created,
calling this.account.connection.prepare_async() will start trying to
connect it. If the host is unreachable, this will result in waiting for
a network timeout before Telepathy can declare failure.
There are two problems here:
• This network timeout may be longer than the quiescence timeout built
in to the IndividualAggregator. There’s not much we can do about that.
• The GError was not previously handled properly, so the PersonaStore
would never reach quiescence anyway.
Fix this by falling back to loading the cache if connecting fails.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
| |
Combine all the simple set-* tests into a single set-properties test,
eliminating a lot of redundant code.
Fix the remaining code to correctly avoid race conditions, as in commit
023a563.
|
|
|
|
|
|
|
|
|
| |
Drastically simplify the tests for retrieving persona details from
Edsf.Personas, by using some of the newer test utility functions in
libfolks. This has the useful side effect of fixing a common race
condition where the test’s individuals-changed callback would be called
during test teardown, and assertions in it would fail because it didn’t
expect any personas to be removed.
|
|
|
|
|
| |
Previously, unit test closures had to be entirely pure — now they can
store data. This will be useful for adding unit tests dynamically.
|
| |
|
| |
|
|
|
|
| |
For debugging timeout problems.
|
|
|
|
|
|
|
|
|
|
| |
Previously, we spawned the EDS source registry process, then created a
.source file to it to use. However, it looks like sometimes it can
notice the file part-way through us writing to it, load it, then fail to
parse it.
Fix that by writing out the file before spawning the EDS source registry
process.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We can’t connect to the objects_modified signal, as we can’t guarantee
whether the handler installed in _commit_modified_property() or the
_contacts_changed_cb() handler will be called first. Since they both
enqueue to the idle queue in order, the callback from
_commit_modified_property() will still operate on an outdated
Edsf.Persona if it is called first.
This race condition can only happen when _commit_modified_property() is
called with a null property_name, which happens for custom properties
and extended fields.
Fix this by listening for changes to the Edsf.Persona.contact property
instead of objects_modified. This is always guaranteed to change when a
Persona is updated (because it represents the current vCard), and fixes
the ordering problem, because the Persona’s properties will have been
updated by the time the contact property is notified (see
Edsf.Persona._update() for details).
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Consistently use TestUtils.loop_run_with_timeout() instead of
TestUtils.loop_run_with_non_fatal_timeout(). Fix a number of race
conditions where (for example) individuals-changed can be emitted as
part of teardown of the test, while the test’s signal handler is still
connected to it, resulting in an assertion failure because the handler
does not expect personas to be removed.
This should also speed the tests up a bit, because they’re no longer
waiting for a timeout before checking for success. They now quit the
main loop on success.
|
| |
|
|
|
|
|
| |
The bindings for file attributes were changed in Vala 0.16 (we depend on
Vala 0.22), so it’s safe to port from the deprecated API to the latest.
|
|
|
|
|
| |
The latest version of valac picks up the omission of the catch block for
generic GLib.Error.
|
|
|
|
|
|
|
|
|
| |
Older version of EDS do not treat the following two calls as equivalent:
new VCardAttribute (null, "X-HELLO")
new VCardAttribute ("", "X-HELLO")
when they should be. Work around that by always passing null anyway.
https://bugzilla.gnome.org/show_bug.cgi?id=751044
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
This reverts commit cdd1ff41c6b58841f3be29845ccab2668e35c0db.
It turns out it’s more pain to disable libsocialweb and try to fight the
irrepressible combination of automake and Vala than it is to just
un-break libsocialweb to get distcheck to pass. Turns out that it’s
basically impossible to stop automake trying to generate C files from
.vala files, even if those .vala files are listed in EXTRA_DIST and not
SOURCES, or are hidden behind an if ENABLE_LIBSOCIALWEB.
|
|
|
|
|
| |
folks-dummy was missing from it; it should have been referencing
$(distclean_dirs) all along anyway.
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
| |
If the user runs folks-inspect and it fails (for example, because the
individual ID they’ve specified doesn’t exist), return a non-zero exit
status, rather than returning zero and pretending everything is perfect.
Interactive mode continues to return zero.
|
|
|
|
|
| |
This seems to have been added in recent versions of Tracker, or just
exposed in the introspected bindings.
|
|
|
|
| |
This fixes a number of compiler warnings.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|