When the remote peer closed the connection, a read notification needs
to always be emitted, otherwise the higher layer does not get the
disconnected signal. From the other side, underlying QAbstractSocket
object could temporarily disable notifications from the engine at
any time. To avoid possible blocking of the socket, take a pending EOF
into account when the read notifications are re-enabled.
Change-Id: Iac9d4e2f790530be3500baf5a2000f1f63df5cc2
Reviewed-by: Ulf Hermann <ulf.hermann@theqtcompany.com>
Introduce QVERIFY2() to get some information when exactly it fails.
Change-Id: Icaddf2ecae434d0bafc90c18458c5ee067dfd506
Reviewed-by: Edward Welbourne <edward.welbourne@theqtcompany.com>
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@theqtcompany.com>
After canceling the asynchronous read operation, the
notified() slot receives ERROR_OPERATION_ABORTED.
We must not handle this situation as an error.
This amends commit 5ce567c5.
Task-number: QTBUG-48336
Change-Id: Iff948ceb3ad1f805a9de8c188fbc39ed4c76ba82
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
Different multicast tests fail on different platforms for different reasons.
Blacklist them to get rid of insignificant and later fix/un-blacklist.
Change-Id: I91548366c7666478ea1cc446bbf337becfdefd49
Task-number: QTBUG-46612
Reviewed-by: Tony Sarajärvi <tony.sarajarvi@theqtcompany.com>
After calling connectToHost(), the socket enters HostLookup state. At this
stage, the socket engine was not created yet, and writing to the socket
should result in either data buffering or an error. So, add a check for
d->socketEngine to prevent a crash on unbuffered sockets.
Task-number: QTBUG-48356
Change-Id: I15ea9ce7de97ce6d7e13e358eca5350745b556bb
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
Reviewed-by: Richard J. Moore <rich@kde.org>
There was a small amount of time between the last readDatagram() call
and disabling a read notifier in case the socket had a pending
datagram. If a new datagram arrived in this period, this qualified as
absence of a datagram reader. Do not change the read notifier state
because it is disabled on canReadNotification() entry and always enabled
by the datagram reader.
Thanks to Peter Seiderer, who investigated the same: "Querying
hasPendingDatagrams() for enabling/disabling setReadNotificationEnabled()
is racy (a new datagram could arrive after readDatagam() is called and
before hasPendingDatagrams() is checked). But for unbuffered sockets the
ReadNotification is already disabled before the readReady signal is
emitted and should be re-enabled when calling read() or readDatagram()
from the user."
However, this patch does not completely solve the problem under Windows,
as the socket notifier may emit spurious notifications.
Task-number: QTBUG-46552
Change-Id: If7295d53ae2c788c39e86303502f38135c4d6180
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This commit changes the readDatagram() and writeDatagram() virtual
functions to take a QIpPacketHeader as meta data, instead of a
QHostAddress/quint16 pair. As previously, the header is an "out"
parameter for readDatagram() and an "in" parameter for writeDatagram().
The header pointer in readDatagram() is allowed to be null if the
PacketHeaderOptions indicates WantNone. Otherwise, it must not be null.
The extra options parameter is introduced because we may not always want
all the metadata upon reception. For sending, we know what to include or
not based on what's set in the incoming header parameter.
QIpPacketHeader splits sender and destination because we'll be able to
return both on datagram reception.
Change-Id: Iee8cbc07c4434ce9b560ffff13ca4213255008c7
Reviewed-by: Richard J. Moore <rich@kde.org>
- Replace Q[TRY]_VERIFY(pointer == 0) by Q[TRY]_VERIFY(!pointer).
- Replace Q[TRY]_VERIFY(smartPointer == 0) by
Q[TRY]_VERIFY(smartPointer.isNull()).
- Replace Q[TRY]_VERIFY(a == b) by Q[TRY]_COMPARE(a, b) and
add casts where necessary. The values will then be logged
should a test fail.
Change-Id: Icaa1edafcc6e2779fbd6dbc2c058544d6e07f1e9
Reviewed-by: Richard J. Moore <rich@kde.org>
Fix warnings:
tst_qtcpsocket.cpp:245:9: warning: private field 'numConnections' is not used [-Wunused-private-field]
int numConnections;
^
tst_qtcpsocket.cpp:1834:9: warning: private field 'exitCode' is not used [-Wunused-private-field]
int exitCode;
tst_qcheckbox.cpp:86:10: warning: private field 'tmp' is not used [-Wunused-private-field]
uint tmp;
^
tst_qcheckbox.cpp:88:10: warning: private field 'tmp2' is not used [-Wunused-private-field]
uint tmp2;
Below warning is caused by code #ifdefed for OS X only, make it a local variable:
tst_qlabel.cpp:114:16: warning: private field 'test_edit' is not used [-Wunused-private-field]
QLineEdit *test_edit;
Change-Id: I53c755545fe2e7ca1f053f40c8c0e50aec2efcdd
Reviewed-by: Mitch Curtis <mitch.curtis@theqtcompany.com>
I'm getting an error with WSARecvFrom in nativeBytesAvailable() and I
don't know why. The number of bytes obtained is correct and the test
works for 2 bytes, so let's use that.
The Windows nativeBytesAvailable() function has a comment saying that
WSAIoctl sometimes indicates 1 byte available when there's a pending
error notification, so that function proceeds to do a WSARecvFrom to
peek the number of bytes. Somehow that function in this test is getting
a SOCKET_ERROR I can't explain.
Change-Id: Ic5b19e556e572a72a9df9a405b1fee3b7efb8b24
Reviewed-by: Richard J. Moore <rich@kde.org>
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 system is no longer in the CI rotation and we haven't had reports
of the same issues happening on later versions. Either the issues have
since been fixed or they were never an issue in Qt in the first place.
This commit has the additional benefit of getting rid of the following
shell error when qmake was run:
sh: line 0: [: =: unary operator expected
as /etc/lsb-release hasn't contained DISTRIB_CODENAME for some time and
proper quoting was never implemented (not even qtcpsocket.pro).
Change-Id: Ia0aac2f09e9245339951ffff13c829e910ee64e9
Reviewed-by: Albert Astals Cid <albert.astals@canonical.com>
Reviewed-by: Tony Sarajärvi <tony.sarajarvi@digia.com>
Reviewed-by: Richard J. Moore <rich@kde.org>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Properly QSKIP tests that use disabled QProcess and symlink
features instead of excluding them silently by #ifdef.
Other reason is that moc doesn't respect QT_NO_* defines
in class definition which causes build issues on some
platforms.
Change-Id: I041020f7452f7d36c7ec8a5866a4ba5eb23d1f94
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
Reviewed-by: Jason McDonald <macadder1@gmail.com>
Other parts of tst_qudpsocket.cpp already did this check.
Change-Id: Iee8cbc07c4434ce9b560ffff13ca545a03c9596a
Reviewed-by: Richard J. Moore <rich@kde.org>
This has been known to be broken for a while. Now it works: you can bind
and you'll retain the port (and the file descriptor) for the connect
call. Incidentally, in fixing the binding for more than one IP for the
hostname (with event loop), this commit fixes the setSocketDescriptor
XFAIL.
[ChangeLog][QtNetwork] Fixed a bug that caused both QTcpSocket and
QUdpSocket to close the socket and lose any bound ports before
connecting. Now bind()/setSocketDescriptor() followed by connect() will
retain the original file descriptor.
Task-number: QTBUG-26538
Change-Id: I691caed7e8fd16a9cf687b5995afbf3006bf453a
Reviewed-by: Richard J. Moore <rich@kde.org>
Actually, it tests the buffered QTcpSocket. Place it in
tst_qtcpsocket.cpp instead of tst_qabstractsocket.cpp.
Change-Id: I3055c4773d0de74c238be4f11b2d1c07ddad4485
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
QAbstractSocket::close() always calls QIODevice::close(), which resets
QIODevice's internal read buffer. So it makes no sense to make same calls
from disconnectFromHost(). This made the closeCalled private member
superfluous.
Change-Id: I4ec64e9711490e44e737763e4ed7fb41bffe2556
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>
The documentation says that QUdpSocket emits readyRead() only for one
datagram and that if you don't read it, the class will not emit again.
That should be implemented by disabling of the socket notifier once we
have the datagram already read, but was broken.
In turn, that breakage caused a live-lock of the event loop: since we
didn't disable the notifier nor read the pending datagram, the event
loop would fire every time for the same datagram.
The re-enabling of the notifier was already working.
Task-number: QTBUG-43857
Change-Id: Ic5d393bfd36e48a193fcffff13bb32ad390b5fe8
Reviewed-by: Peter Hartmann <phartmann@blackberry.com>
Reviewed-by: Richard J. Moore <rich@kde.org>
Linux gracefully allows us to do that and treat the v6 socket as if it
were v4. Other OS (notably OS X) aren't so forgiving.
Change-Id: I13dd3274be2a4b13e8b1eef93cbc2dd17b648f96
Reviewed-by: Richard J. Moore <rich@kde.org>
We want to use "localhost" if the server's address is "any", as some OS
can't send datagrams to "any" (e.g., OS X and FreeBSD).
Change-Id: I1004bc2282e7f930cdb7ed394aa9f4b5a1cfcf82
Reviewed-by: Richard J. Moore <rich@kde.org>
On my Mac Mini, port 5000 is in use, which means the broadcasting test
fails.
Change-Id: Ifb0883263e277f388342430349ea7315d42f324a
Reviewed-by: Richard J. Moore <rich@kde.org>
It was unconditional. Someone forgot to check for IPv6 support before
skipping IPv6 tests.
Change-Id: I7b11528ad02560f0db9defde3c64f76f48a6c1f8
Reviewed-by: Richard J. Moore <rich@kde.org>
QUdpSocket doesn't support binding to QHostAddress::Any and then joining
an IPv4 multicat group since QHostAddress::Any is really an IPv6 socket
with v6only = false. The test did check this case, but failed to ignore
the warning.
Change-Id: I62d782408319a6e566e0ff1a6081b706ac1f669c
Reviewed-by: Richard J. Moore <rich@kde.org>
With IPv6, you cannot bind to a multicast address. You need to bind to a
local address only. The previous tests either checked this or didn't
check the result of bind().
Change-Id: Ief70887d8988fc1bc4394cf6ff34b5d560e5748e
Reviewed-by: Richard J. Moore <rich@kde.org>
Sending 100*8 packets of each type of message is WAY overkill. That's a
stress test without limiting. My Linux system starts reporting EAGAIN on
the socket, so reduce the amount of data sent.
Change-Id: I153f44cf3b91d37526dac580b400114cc80b1769
Reviewed-by: Richard J. Moore <rich@kde.org>
IPv6 has no such thing, so don't try to bind to an IPv6 address to send
broadcasts (even though that works) and it's a poor idea to bind to IPv6
to receive broadcasts. Moreover, skip any IPv6 network addresses
(broadcast() is invalid).
Change-Id: I2829b042c000158565adfd92db682f37d67dacae
Reviewed-by: Richard J. Moore <rich@kde.org>
If you don't have /etc/lsb-release, you'd get
sh: line 0: [: =: unary operator expected
Change-Id: Idb5c79f799879e4d32cd640ef74fb388227f831e
Reviewed-by: Richard J. Moore <rich@kde.org>
It doesn't make sense because there is no command to ask the proxy
server to join a multicast group. At best, we could write a datagram via
proxy without joining, but we definitely can't receive.
Change-Id: Icc6b54572a053fb7821dfca1f4111f2046ff8686
Reviewed-by: Richard J. Moore <rich@kde.org>
