This reverts commit 618e2cc081. The
original commit has a section of code that I failed to review properly
and is of questionable functionality.
Change-Id: I61c53d7b8b2aa7c3312292b017a18aba7da11bc5
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This patch includes setup of class member 'msg' in
QDBusMessagePrivate::toDBusMessage() to be able to get the
serial after message sending.
Testcases for comparing the 'reply serial to' with the 'serial'
are included.
Task-number: QTBUG-44490
Change-Id: Iae7c48f5b0c70a6c5ae500904072b38b46dfd876
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This often happens in applications. Besides, we are expecting at least a
call to RequestName to happen.
Change-Id: Ifd2454ffba454fd591d0ffff1425a84563267d19
Reviewed-by: David Faure <david.faure@kdab.com>
To retain a bit compatibility with applications developed in the last 9
years that expect that QDBusConnections won't process their events until
the event loop runs, we now suspend the handling of incoming messages
in the two default buses (and only in them) and resume when the event
loop starts. This is required because the new threaded QtDBus would
otherwise process incoming messages that the application didn't expect
it to.
For example, if the application first acquires names on the bus and only
after that registers objects with QtDBus, there's a small window in
which the name is acquired and visible to other applications, but no
objects are registered yet. Calls to those objects may be received,
would then be processed in the QDBusConnectionManager thread and fail.
The work around is to disable the actual handling of method calls and
signals in QDBusConnectionPrivate::handleMessage. Instead, those
messages are queued until later.
Due to the way that libdbus-1 works, outgoing method calls that are
waiting for replies are not affected, since their processing does not
happen in handleMessage().
[ChangeLog][Important Behavior Changes] QtDBus now uses threads to
implement processing of incoming and outgoing messages. This solves a
number of thread safety issues and fixes an architectural problem that
would cause all processing to stop if a particular thread (usually the
main thread) were blocked in any operation. On the flip side, application
developers need to know that modifications to a QDBusConnection may be
visible immediately on the connection, so they should be done in an
order that won't allow for incomplete states to be observed (for
example, first register all objects, then acquire service names).
Change-Id: I39cc61d0d59846ab8c23ffff1423c6d555f6ee0a
Reviewed-by: David Faure <david.faure@kdab.com>
Use QVERIFY2() with QTemporaryDir/File::errorString() consistently.
Attempt to catch issues like the below warning and follow-up issues.
QSYSTEM: tst_QFiledialog::clearLineEdit() QFileSystemWatcher: FindNextChangeNotification failed for "C:\Users\qt\_____aaaaaaaaaaaaaaaaaaaaaa" (Access is denied.)
Task-number: QTBUG-47370
Change-Id: I58a6e87c502627e976efa62ad73c912f3b2d49fa
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@theqtcompany.com>
The explanation is in the code comment. Ever since QDBusConnections
began being processed in a separate thread, we were relying on the fact
that the main thread didn't begin processing its event queue until the
second event got posted (the event loop only exits after it has finished
processing all pending events). We had a race between the main thread
starting its processing, at which point it decides which is the last
event it will process, and the QDBusConnectionManager thread posting the
second event.
This is very fragile code, since it depends on the behavior of
QDBusConnectionPrivate (how it stores the signal relays in a hash) and
that of QHash with duplicate keys. This only works because the hash
key between the two connections is the same (it's only dependent on the
method name and interface name). If we ever begin using something that
isn't the same between "control" and "p", then with QHash's randomness,
we'll be racy again.
Change-Id: I42e7ef1a481840699a8dffff1406c3a4674ec3a6
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
In some cases it's not possible to use QT_HASH_SEED, specially when
we need to set the environment variable from inside the application,
as dynamically loaded libraries or plugins may create static QHash
instances. That would set qt_qhash_seed to a value different from
-1 and skip the env var value.
For those cases, and when we still want to set qt_qhash_seed, we
provide a way to enforce its value.
Auto-tests accessing qt_qhash_seed directly have been updated
accordingly. Usage in qdoc, uic and rcc has been left as is
for the time being.
Change-Id: I3b35b4fa0223c83b1348a6508641905a2a63266f
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
P2P connections don't have senders and receivers, so asking
QDBusConnection to connect to a signal with a sender was a mistake
(added in 5368e44a86). Due to an internal
bug, this never presented itself -- double fault.
Fix the connection so that we don't get unit test failures when the bug
is solved.
Change-Id: I9a75ad8521ae4e5cbbe5ffff13d1a78b7dea6d07
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
With kdbus, we won't have a regular signal, but instead a special
message. So keep the logic of what to do in QDBusConnectionPrivate.
The #ifdef is to make sure the bootstrapped qdbuscpp2xml continues to
build in cross-compilation environments.
Change-Id: Iee8cbc07c4434ce9b560ffff13d06f0d9904cb6d
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
Now we know that all timers and socket notifiers get created only in the
QDBusConnectionManager thread.
Incidentally, this reduced code duplication.
Change-Id: I27eaacb532114dd188c4ffff13d5075a8d2efb0b
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
Each application will have one thread dedicated for this, for all
QDBusConnections. I wouldn't mind sharing such a thread with other uses
in Qt, provided none of them ever block (the QProcessManager thread
comes to mind, but it's going away soon).
The cost associated with this change in this commit is so far rather
minimal. All incoming D-Bus calls need to be handled after an event is
posted anyway, to avoid deadlocking on reentering libdbus-1 functions
that acquire locks still held. The cost is the one more thread running
and the cost of synchronizing them when an event is posted.
The benefits far outweigh that cost: no longer will we have problems of
QtDBus failing to run if the main system or session connections are used
before QCoreApplication is run. Moreover, events can be received and
handled in aux threads even if the main thread is blocked on some
operation.
Note: this commit may not be testable (tst_qdbusconnection may fail)
Task-number: QTBUG-43585
Change-Id: Ic5d393bfd36e48a193fcffff13b737556ccd11a8
Reviewed-by: Albert Astals Cid <aacid@kde.org>
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
This simplifies the code a little by having a single code path. More
importantly, we no longer need to call the evil function
dbus_connection_send_with_reply_and_block. That function acquires a lock
on the socket transport inside libdbus-1, which means all threads need
to wait until the one call gets unblocked before they can continue.
To do that, this commit reimplements the QDBus::Block part of
QDBusConnectionPrivate::sendWithReply by reusing the existing call to
sendWithReplyAsync() and then doing a blocking-wait with
QDBusPendingCallPrivate::waitForFinished().
By using (Q)DBusPendingCall and the threaded connection approach (next
commit), now we never block on the socket. That also means the code to
call dbus_pending_call_block() is no longer necessary and the
waitForFinished() function itself can be considerably simplified.
As a side-effect of no longer blocking, a number of pre-existing race
conditions that used to be hidden showed up.
Note: this commit deadlocks without the threading (next commits).
Task-number: QTBUG-43585
Change-Id: Ic5d393bfd36e48a193fcffff13b73754954a3f7d
Reviewed-by: Albert Astals Cid <aacid@kde.org>
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
Otherwise, if I open tests/auto/dbus/dbus.pro in Qt Creator, it shows me
"test", "test2", "test3", "test4" and it's very hard to know which test
is which.
Change-Id: Iee8cbc07c4434ce9b560ffff13d0654696c025b7
Reviewed-by: Frederik Gladhorn <frederik.gladhorn@theqtcompany.com>
This isn't specific to an Ubuntu version, unfortunately. It also fails
on OpenSuSE occasionally and other Ubuntu versions.
Change-Id: I6a1ca55a198270f1a1e8a9916e9f768762211550
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
They didn't show up in the "old" CI runs because they usually pass the second
time they are executed - which the testrunner does. The new CI doesn't do that
anymore, instead we now mark those tests explicitly and will track their record
of passing and failing in the new metrics database.
Change-Id: Id34dd6f792f38995b07b6fec88f833df64de2f8b
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
This function was introduced alongside the support for Unix file
descriptors, so it's a good indicator of whether Unix FDs are
supported. Ever since dbus_minimal_p.h, however, DBUS_TYPE_UNIX_FD may
be defined even if the system libs don't support it.
In order to fix this issue, I had to fix what was apparently a merge
conflict resolution mistake and remove the #ifdef around the test. Doing
the latter is a good idea due to moc being unable to find <dbus/dbus.h>.
This was tested with both linked and dynamically-loaded libdbus-1.
Task-number: QTBUG-46199
Change-Id: I66a35ce5f88941f29aa6ffff13dfb4b5438613a3
Reviewed-by: Jani Vähäkangas <jani.vahakangas@theqtcompany.com>
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
Because if it doesn't, then calling dbus_type_is_fixed or is_basic may
result in a failed assertion.
process 16304: arguments to dbus_type_is_fixed() were incorrect,
assertion "_dbus_type_is_valid (typecode) || typecode ==
DBUS_TYPE_INVALID" failed in file dbus-signature.c line 345.
Change-Id: Idf715b895bac4d56b4afffff13db2ed71b1516a5
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@theqtcompany.com>
Commit 748abf9347 changed the message, but
the CI either did not run the QtDBus tests when integrating or it
ignored the results.
Task-number: QTBUG-45317
Change-Id: Iee8cbc07c4434ce9b560ffff13d058f21b73ac05
Reviewed-by: Liang Qi <liang.qi@theqtcompany.com>
Reviewed-by: Alberto Mardegan <mardy@users.sourceforge.net>
Tests are now passing in CI.
Change-Id: I0051fb7070c1c1027c557eba9dde6367ad59ac7a
Task-number: QTQAINFRA-837
Reviewed-by: Frederik Gladhorn <frederik.gladhorn@theqtcompany.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The QDBusMessage::createSignal() static method doesn't take a parameter
for specifying the destination service of the signal. While this is not
a widely used feature, it can be useful to avoid waking up all connected
clients when the service knows what are the clients which are interested
in the signal.
This commit adds a QDBusMessage::createTargetedSignal() method which
also takes the destination service as its first parameter.
Change-Id: I9fdca53673a6944c39c93c1efd69a9d02859684e
Task-number: QTBUG-44704
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Currently QDBus relies on a key in QMetaClassInfo to understand the DBus interface name.
This patch allows QDBus to also use a specified interface name in the registerObject call
instead of relying on QMetaClassInfo that might not be there (if the QObject was created
in QML or Javascript for example).
Change-Id: Ie02b2c67e7deb07f43e35eb166c11833fcbf38f3
Task-number: QTBUG-44074
Reviewed-by: Kevron Rees <kevron.m.rees@intel.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Qt copyrights are now in The Qt Company, so we could update the source
code headers accordingly. In the same go we should also fix the links to
point to qt.io.
Outdated header.LGPL removed (use header.LGPL21 instead)
Old header.LGPL3 renamed to header.LGPL3-COMM to match actual licensing
combination. New header.LGPL-COMM taken in the use file which were
using old header.LGPL3 (src/plugins/platforms/android/extract.cpp)
Added new header.LGPL3 containing Commercial + LGPLv3 + GPLv2 license
combination
Change-Id: I6f49b819a8a20cc4f88b794a8f6726d975e8ffbe
Reviewed-by: Matti Paaso <matti.paaso@theqtcompany.com>
isnan is C99 and POSIX.1, which the older MSVC do not support. Use the
Qt equivalent.
Change-Id: Ic5d393bfd36e48a193fcffff13b8679cb83d12db
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
Two serious mistakes:
- we need to call dbus_server_free_data_slot as many times as we call
dbus_server_allocate_data_slot
- we need to delete the d pointer...
The changes to the unit tests are simply to cause the used peer
connections to be removed so they don't show up in valgrind.
Change-Id: I9fd1ada5503db9ba481806c09116874ee81f450d
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
This has been deprecated since QDBusContext was introduced (Qt 4.3). So
it's time to remove the functionality.
[ChangeLog][Important Behavior Change] QDBusConnection::sender()
(deprecated since Qt 4.3) has changed to always return an invalid
QDBusConnection. To know what connection the incoming call was received
from, use QDBusContext.
Change-Id: I355efb82c14e54ed718c8f892d8267e727b19118
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
There's a change in Qt 5.4.0 that makes Qt compile with its own set of
D-Bus headers, which means QT_CFLAGS_DBUS may be empty. Thus, we can't
compile or link if we're using the actual libdbus-1 API to build the
test.
This commit makes these unit tests use the same dynamic loading
mechanism.
Change-Id: I56b2a7320086ef88793f6552cb54ca6224010451
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
The error of "Not connected".
This incidentally solves a crash when QDBusServer().lastError() is
called but libdbus-1 couldn't be found.
Change-Id: Id93f447d00c0aa6660d4528c4bbce5998d9186a8
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
On the unit test side, everything is sequential: we first ask for the
connection, verify that it is connected, then ask the remote side via
the session bus if it is connected. Unfortunately, the remote site may
handle things in a different order: it may handle the incoming function
call to "isConnected" before doing accept(2) on the listening socket.
So, instead, make the local side block until the connection is received
on the other side. On the remote, we don't block, instead we use the
feature of delayed replies.
Change-Id: Ie386938b8b39dd94a9d7e5913668125fb4a3c7da
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
So we can get the output when running with QDBUS_DEBUG=1.
Change-Id: I6a6b8e0d82727c522914fb90a7ce594c86307d8f
Reviewed-by: Frederik Gladhorn <frederik.gladhorn@theqtcompany.com>
Instead of killing them outright (and note that terminate() doesn't work
on Windows), ask them nicely to exit on their own. This way, if we run
them in valgrind, valgrind gets a chance to print the leak check output
and summary.
Change-Id: Ib6cc8d4560ff0bf255f94980eb220e97592c00f0
Reviewed-by: Frederik Gladhorn <frederik.gladhorn@theqtcompany.com>
Wait for the subprocess to print "ready" before assuming that it is
ready to receive calls. waitForStarted() will return as soon as the
child is running, but it may not have registered on D-Bus yet.
This also solves the synchronization problem more elegantly than how
tst_qdbusmarshall.cpp was trying to do it.
Change-Id: I548dfba2677cc5a34ba50f4310c4d5baa98093b2
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
The executables are not in the same dir as on Unix, so we need to use
QFINDTESTDATA to find them. The DESTDIR setting prevents qmake from
placing the executables in a "debug/" subdir.
Change-Id: I1d6d10e6f6f109f55fd9809dcf83da0386f38772
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
Rohan was right in e88f9a92b7 to stabilize
the test and reset the state, but killing the subprocess is overkill.
All we need is to reset the state in both applications, which includes
disconnecting and reconnecting to the peer, to discard any sent but not
yet received messages.
Change-Id: Ie01392e6e63bd70ef8345217d3fc641ed63c7aba
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
QDBusServer::address() will return an empty QString, which caused the
tests to fail later with no apparent reason.
Change-Id: I86f448dfc67a6cdb27ecda2d490f335766cfaf4f
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
This allows the tests to be run on Windows too by using TCP socket
connections instead of requiring Unix sockets. The tests shouldn't have
hardcoded the path, which came from QDBusServer anyway. Now the tests
simply defer to QDBusServer.
This is a slight behavior change for Windows, but not one that should
matter since anyone who was using the default constructor resulted in a
QDBusServer that failed to listen.
[ChangeLog][QtDBus][QDBusServer] Fixed a bug that made QDBusServer's
default constructor try to bind to a Unix socket on non-Unix systems.
Now QDBusServer will attempt to bind to a TCP socket instead.
Change-Id: I2a126019671c2d90257e739ed3aff7938d1fe946
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
When generating the DBus signature of a registered custom type the
marshaller appends the signatures of the map entries and array items
after the map/array causing an invalid DBus signature to be generated.
This happens because beginArray() and beginMap() output the full
signature of the data.
Fixed by suppressing changes to the signature within
beginArray()/endArray() and beginMap()/endMap() blocks.
Change-Id: Icaf23b2fe58a3e1f575b81e4a100f02684f68452
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Most of QtDBus already needs very little from libdus-1, so create an
extra header containing the minimum API we actually need.
One large advantage of this solution is that now QtDBus can always be
enabled, even if the system doesn't have libdbus-1 installed. This is
interesting on OS X, where libdbus-1 is often installed by Homebrew or
MacPorts, which may include extra libraries we don't want in our
packaging.
Change-Id: I1b397121ec12eeca333ef778cf8e1c7b64d6b223
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Several of the unit tests request that the peer emit more than one
signal, but only handle one. The rest of the signals stay queued in the
socket and will be delivered at the next test, causing it to fail often.
This doesn't happen in the tests with the bus. There, we don't receive
the extraneous signals due to AddMatch/ReceiveMatch on each signal
individually and the synchronous nature of the emission (the signals
have already been emitted by the next AddMatch and cannot match it).
Task-number: QTBUG-42145
Change-Id: I743a0553074972042fca46b76db5d9e7b3209620
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@digia.com>
