| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
| |
Change-Id: I58cf61540bc53e314988753a252cbc71003ce581
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Even though this has been introduced in 5.1, officially, we can rely on it in
the previous versions even if private API that it was.
* Global variables are better avoided.
* Q_GLOBAL_STATIC is thread-safe in Qt 5.
* QLibrary depends on QCoreApplication and we need to make sure that is
initialized afterwards, respectively.
The instantiation is now moved into the corresponding source file so that if it
is included at more than one place, it will not be instantiated each time it is
used in a new source file including this header.
Task-number: QTBUG-36870
Change-Id: I96de1257e5836b69d0c48b717d7c2e708d6b0fee
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This makes the operation thread-safe. It also helps with preventing the leaks
for child processes. Also, there is now some protection against EINTR when the
system call has to be restarted.
Thanks go to Thiago Macieira for providing the original patches.
Task-number: QTBUG-36824
Change-Id: I90cac87c5e1c55a57e5ba6c20313446c43c6d242
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|
|
|
|
|
|
|
|
| |
Task-number: QTBUG-36865
Change-Id: I332901ad0118f1176b0d3b8e45c8000f0dfb59ee
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The serial port very slow device therefore the system
call write() doesn't guarantee that data were transferred
completely to the device. It guarantee only that data were
sent to the driver from the user space.
Thus it is wrong to emit the bytesWritten() signal at once
after successful writing.
The correct solution is to store the number of transferred
data in some variable (e.g. pendingBytesWritten in this
case) and to wait for the next notification from the
WriteNotifier that the write FIFO of device is empty (i.e.
when all data were really transferred) and ready to
transmission of the next portion of data.
Also good decision is to divide a data transfer operation
into two methods: startAsyncWrite() and completeAsyncWrite(),
similar with the Windows implementation.
Where the startAsyncWrite() invokes the write() system
call, and the completeAsyncWrite() is invoked by a
writeNotifier to complete writing operation or start a new
startAsyncWrite() operation.
Tested on Arch Linux 64 bit with the on-board and the USB
(PL2303) serial ports with use Qt4 and then Qt5.
Change-Id: I1c274b2052a9bd54811586c6f1cfdf080b400263
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Revert "Open file descriptors thread-safely on Unix and don't leak them"
This reverts commit 691212d9e492b12590d5b7f4e2b24921911d4b17.
Revert "Protect against EINTR in Unix non-atomic I/O calls"
This reverts commit a8597cbae47076e33b638c2e593f852c8c0a02d5.
Change-Id: I6d931ae6d3b4bee9be019fdf4a08ecc5a41a2b0f
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The device registry property of the SPDRP_HARDWAREID is not
used any more. This property was used earlier for parsing of
the VID/PID and had the type of the REG_MULTI_SZ which was
interpreted as QStringList, see commit:
d8dc10efb1714dcfafa6a08e107fd31fc1e3ce0e
The current code is using only string based concept, hence
changing the interface to QString.
Tested on Windows 8 with on-board and virtual com0com serial
ports using the cenumerator example.
Change-Id: I31e4f85f1b145021c1b34b81ba890553604a0531
Reviewed-by: Laszlo Papp <lpapp@kde.org>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|
|
|
|
|
|
|
|
| |
read(), write() and close() may return EINTR, indicating the system call
should be restarted. The qt_safe_xxx versions of those functions do it.
Change-Id: Ifef3b98131a7e24a5defea3e3af9345a505ad733
Reviewed-by: Laszlo Papp <lpapp@kde.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
::open directly is not thread-safe. You need to pass O_CLOEXEC if the OS
supports it. That's the only thread-safe way to open files. qt_safe_open
does that for you.
This patch also implicitly makes all serial devices be opened with
CLOEXEC on all systems, which prevents the file descriptors from leaking
to child processes.
Change-Id: I83a2657fa8d9ccca1b42182bcd83416091066045
Reviewed-by: Laszlo Papp <lpapp@kde.org>
|
|
|
|
|
|
|
|
|
| |
The cfmakeraw() function is an extension to POSIX and thus not fully
portable. It doesn't exist on Solaris and Illumos.
Source: http://www.perkin.org.uk/posts/solaris-portability-cfmakeraw.html
Change-Id: Icf514fe32c01e85baff2a0d49321e45a320703b7
Reviewed-by: Laszlo Papp <lpapp@kde.org>
|
|\
| |
| |
| |
| |
| |
| | |
Conflicts:
src/serialport/qserialport_win.cpp
Change-Id: Ib00a645c6660a3c81dbc9e159110f7bfc392ea36
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
They are not much of use due to the low-level OS limitations as of now, hence
the obsolete proposal. It can be brought back to life at any point when we have
a more clear idea what to do with this API.
"faking" this signal as RealTterm probably does it, is not a nice impression
for the end users. That would be more like a software "signal" (which we
already have in place) rather than a real pinout hardware signal.
It is interesting why Windows, Linux, etc do not provide an option for this.
There must be some strong reason for it.
Change-Id: Ia065e2ee9226d16e724f5e2690b25b954329d78e
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The documentation contains broken links due to this, and it is also inconsistent
with the rest. Properties should be referred to as properties, not individual
methods like setter, getter, notified signal, etc.
The change is tested with "make docs" and the broken links are history.
Change-Id: Id94cba1f88503754318c865793229e3d3947bec6
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Jerome Pasion <jerome.pasion@digia.com>
Reviewed-by: Martin Smith <martin.smith@digia.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
This change eliminates the following warnings when using "make docs":
/home/lpapp/Projects/qt/qt5/qtserialport/examples/serialport/doc/creaderasync.qdoc:28:
warning: Cannot find file 'creader/creader.pro' or 'creader/creader.qmlproject'
/home/lpapp/Projects/qt/qt5/qtserialport/examples/serialport/doc/creaderasync.qdoc:28:
warning: EXAMPLE PATH DOES NOT EXIST: creader
It is also visible that the example is unreachable on the following
documentation page:
http://qt-project.org/doc/qt-5/examples.html
Change-Id: I09f1532fcff6b1f7bfc5c4857bd14252aceb2eb3
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
This is likely to be happening due to the mismatching function signature include
the return type between the header and source file. Two things are being fixed:
1) Dropped scope for the Handle just like with the winId case in QWidget.
2) "const" qualifier is now added for the method.
Here you can see the documentation that it is empty:
http://qt-project.org/doc/qt-5/qserialport.html#handle
Change-Id: I5e9e0732cfd746a863de523fe90e58eafa42c081
Reviewed-by: Alex Henrie <alexhenrie24@gmail.com>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Martin Smith <martin.smith@digia.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
It is necessary to compare size of the readBuffer before and after
completion of each read operation. If size of readBuffer are equal
it means that are no more data available for reading. In this case
the method should returns true if an initial size of readBuffer is
not equal to the current size of readBuffer. Otherwise in all other
cases should be continue waiting until timeout has been expired.
Also must not to do check for the NoError code and to hope for valid
value of this code inside waitForReadyRead(). Because a last error
code do not clears automatically. It lead to false returns of this
method in case the error number is not equal to NoError (for example
it error remained after failed of any previous method). This check
should be implemented in a different way in case of need. But at
present this check is unnecessary because result of the waitAnyEvent()
cover it.
Task-number: QTBUG-33987
Change-Id: Ic8d8e3806fd4863c2720ffb83d5c19eae54d57f0
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Laszlo Papp <lpapp@kde.org>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Updating of the DCB structure (in case of setup of some
properties through setXX() methods) leads to reset of a
status of the DTR line to an initial state.
For example, after opening of port the DTR state was "high",
further the call of method setDataTerminalReady(false)
will drop the DTR to "low" state. But if now to change any
properties of the device (e.g. to setup a new baud rate),
the DTR will again be set to "high".
The reason of this behavior is the fDtrControl flag of
DCB structure:
http://msdn.microsoft.com/en-us/library/windows/desktop/aa363214%28v=vs.85%29.aspx
To control of DTR is used the EscapeCommFunction() function
which has no relation to the DCB structure. Therefore any
re-initialization of the DCB discard the results of the
EscapeCommFunction() function in case a state of the
fDtrControl flag not equal to the real DTR state.
Solution is change of a fDtrControl flag after successfully
call of the EscapeCommFunction() function. Then further change
of any property will be with a correct DCB.
Besides, it is necessary to drop the flag DTR_CONTROL_HANDSHAKE
at opening of device. Because in this mode the manual control
of the DTR line is forbidden.
Thanks to Dr. Alexander W. Lenhardt
Task-number: QTBUG-36490
Change-Id: I14b040761f7e28108db87e667eb76f559be436cb
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Please refer to the MSDN documentation for details:
http://msdn.microsoft.com/en-us/library/windows/desktop/aa363214%28v=vs.85%29.aspx
and
http://msdn.microsoft.com/en-us/library/windows/desktop/aa363201%28v=vs.85%29.aspx
Change-Id: Id60f6d1b4a9237c2ae74063ac52221f42761b004
Reviewed-by: Laszlo Papp <lpapp@kde.org>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Conflicts:
src/serialport/qserialport_win.cpp
Change-Id: I786879a485e78e3637d8cb474f86e3bd406bd237
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
* Windows uses the term handle instead of descriptor.
* It is also consistent with the rest how Qt manages it.
Conflicts:
src/serialport/qserialport_win.cpp
Change-Id: I68a672ea2d8e88e26ad5d822b157ccab79a3d2d9
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
| |
| |
| |
| |
| |
| | |
Change-Id: Iafab6ccc376da7a856b409017bbc42c36c016253
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|
| |
| |
| |
| |
| |
| | |
Change-Id: If8dee1819d493fa4b27eba99de611c61bfdfe7a3
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
When using some software that provide an virtual serial
ports functionality is impossible to use the Win32 SetupAPI
feature to enumerate these serial ports.
For example, such software is "Virtual Serial Ports Emulator"
from Eterlogic: http://www.eterlogic.com/Products.VSPE.html
The only possible simple way to detect these serial ports
it to read the registry key HARDWARE\\DEVICEMAP\\SERIALCOMM:
http://msdn.microsoft.com/en-us/library/windows/hardware/ff546502%28v=vs.85%29.aspx
In this case it is possible to get only a name of the serial
port without any additional information.
This method is simply addition to the main SetupAPI functionality.
Task-number: QTBUG-36526
Change-Id: Ib505aee66b74b6a8ebe16cf88c3060c8267397a0
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
| |
| |
| |
| |
| |
| | |
Change-Id: Ie6b17d95b83ad751130305a6d44cdead3cd46da6
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The current algorithm takes a name of the serial port from
the PortName property which is in the Registry. This value is
filled out by the Ports class installer automatically:
http://msdn.microsoft.com/en-us/library/windows/hardware/ff546514%28v=vs.85%29.aspx
But at using the "AGG Software" software this property is
absent. Therefore the algorithm ignores these serial ports.
The simplest workaround is use of the PortNumber property
which identifies a port number for this software. In this
case a port name defines indirectly, by adding a port number
to the "COM" suffix.
This does not influence other serial ports which are defined
through PortName property.
Tested on Windows XP and Windows 8 with the cenumerator example.
Task-number: QTBUG-32774
Change-Id: I8cda3ed992ff80742511a2952d3fb7e8ac6edc81
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
* handle() is duplicated since the description explains that method.
* It is pointless to put the "object" into the sentence.
* It is consistent to start the sentence with the "if".
Here goes a typical example that is worth following with consistency:
"If successful, returns true; otherwise returns false."
Change-Id: I1c7648237c75381b8452b8c47a59445fff58ad34
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
* The writeSequenceStarted shall be set to true after a successful call
of WriteFile() and to be dropped to false in case of write completion.
* The method startAsyncWrite() shall not call WriteFile() if the
writeBuffer is empty or writing sequence is already started. Also now
there is no sense to check a validity of pointers on data in writeBuffer
before call of WriteFile() because writeBuffer can not be empty.
* It is also expedient to free the writeBuffer in _q_completeAsyncWrite()
method only if numberOfBytesTransferred is not zero.
Tested on Windows 8 with the on-board and the USB (PL2303) serial ports.
Change-Id: Ib70276705fcc3f2396bf7dd6cf6977b62382a64b
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
If the user tries to open the serial port with empty
port name, Windows returns this error code.
The error will be exposed as
QSerialPort::DeviceNotFoundError from now.
Tested only on Windows 8 (32 bit) with on-board serial
port as well as an usb dongle (PL2303) using Qt4 and Qt5.
Change-Id: I9a56a7e51dbdab684b6d2a9fdf6398d8d5018e1c
Reviewed-by: Laszlo Papp <lpapp@kde.org>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| | |
Fix some spelling and grammatical errors in comments that show up in Qt documentation. No changes to code.
Change-Id: I185458e162d115eac25326d239b851b5e14ed5cf
Reviewed-by: Laszlo Papp <lpapp@kde.org>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
This is needless because it is only used once, right after the creation.
Moreover, we would probably need to separate the two calls later into distinct
introspection conditions to give a bit more correct warning whether the sysfs is
present, or it is just not readable for some reason.
It is also consistent with some other places.
It also eliminates the needless indentation level for that relatively long
block.
Change-Id: Ib66e3033c2439315a8034101129cadd2dcdc29d0
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
| |
| |
| |
| |
| | |
Change-Id: Ic1184214f9ce933fc3df47629e0a707e0b8d63f5
Reviewed-by: Laszlo Papp <lpapp@kde.org>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Earlier for writing in writeBuffer was used the
QSerialPortPrivate::writeToBuffer() method which contained
a similar code for each platform.
Therefore it is reasonable to delete this method, and to move
its common functionality to the QSerialPort::writeData(). Also
all platform-dependent code of start of data transfer move to
the QSerialPortPrivate::startWriting() method.
Change-Id: I1423723fd69c05df974f8ba597e3dd71d5a797dd
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
It makes sense to unify internal methods of asynchronous I/O where to each
method of start of operation corresponds the method of completion of
operation:
* startAsyncCommunication() - it is a new method which has a code relating
to the handling of the WaitCommEvent() function. It implementations is taken
directly by copy/paste with minimal modifications relating with adding of
returns a boolean values.
* _q_completeAsyncCommunication() - it is an existing method which is renamed
from the _q_canCompleteCommunication().
* startAsyncRead() - it is an existing method without modifications.
* _q_completeAsyncRead() - it is an existing method which is renamed from
the _q_canCompleteRead(). Earlier this method contained a subfunction of
the completeAsyncRead(DWORD) which is deleted now, and its code (related with
the policy processing) is moved into body of the _q_completeAsyncRead() with
the minimum modifications which do not change it behavior.
* startAsyncWrite() - it is an existing method without modifications.
* _q_completeAsyncWrite() - it is an existing method which is renamed from
the _q_canCompleteWrite(). Earlier this method contained a subfunction of
the completeAsyncWrite(DWORD) which is deleted now, and its code is moved
into body of the _q_completeAsyncWrite() without modifications.
Also in addition is added the new method emitReadyRead() and the code of
policy emulation is moved into new method emulateErrorPolicy().
Change-Id: I58345e3270d676879a16efc4b7f35f74869894d8
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
| |
In case of handling of the TimeoutError it is necessary to check on
existence of the read data (for this concrete implementation of the
example). Otherwise the application always return the TimeoutError
even if a data was successfully read.
Change-Id: I2437461f2ed2806bb7d7e16deb6925df035f62a1
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
|
|
|
|
|
|
|
|
|
| |
Received data should be accumulated at each triggering of readyRead()
during wait timeout. Otherwise application may print out only last
portion of received data.
Change-Id: Ib4d72f70e333298bae31c0eb95d7639b5c4d177d
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A feature of writing of the buffer by chunks of 512 bytes (by default)
was introduced earlier as the developers option in order that it was
possible to regulate load of the CPU and quantity of emitted signals of
bytesWritten() just change WriteChunkSize value in source code.
As shows of the practice, the most optimal mode is writing of whole
content of the buffer without division into chunks. It reduces load of
CPU (because reduces a quantity of the events triggers) and also
simplifies the source code.
But potentially there can be negative consequences of writing of whole
buffer directly in case of broken driver:
* A kernel panic (or BSOD) in case of writing a data large than
driver's internal queue.
* Can be transferred not all data from the buffer, but only their part.
Change-Id: I0fcac1ccf3c752579978b4745771accbf8274267
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* actualWriteBufferSize variable - it is a wrong old heritage related with
the implementation of a code with the analogy from the the QWindowsPipeReader
(where actualReadBufferSize is present).
* acyncWritePosition variable - it is a old heritage when the write notifier
could operate with a multiple writing operations at the same time. For this
purpose it is necessary to know the following write position in the ring buffer.
But currently all write operations happen sequentially, i.e. forbidden to carry
out startAsyncWrite() several times without completeAsyncWrite() completion.
Change-Id: I25b665b638649e6002fb194babfaa73f5ec8a6fa
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The principal idea of this patch - to simplify a code and to minimize
risks of UB in operation with the notifiers.
Details are below:
1) Use a separate instances for the each notifier.
It allows to get rid from an lookupXX() methods because now we always
know which of notifiers was triggered.
2) Now each notifier is created once and used during serial port life.
Moreover now in the open() and the close() methods are carried out a
simply enabling and disabling of the notifiers.
It allows to simplify code and to reduce overhead costs of check of
a validity of the notifier pointers.
3) Use the signals/slots mechanism for the internal events handling.
Use a slots excludes any UB in the notifier handlers because now isn't
present a deep callbacks and so on. Even if some notifier will be deleted,
we never got a crash at processing of triggered handle, we can just got
an some error code and handle it.
4) There is no more need for existence of separate classes of notifiers
which were inherited from the QWinEventNotifier.
Now, the objects of QWinEventNotifier are use directly, only for delivery
a triggered events to the appropriate slots. All logic of a events
handling are in the _q_xx() slots.
5) The overlappedPointer() methods aren't necessary to us anymore.
Now we know each OVERLAPPAD structure instance. It gives us to pass a
pointers of these structures directly to the
ReadFile/WriteFile/WaitCommEvent functions and have direct access to
the hEvent field for the waitForXXX() methods.
6) The errors handling inside of the _q_xx() handlers are simplified.
Now is not used a dptr-pointer to the QSerialPortPrivate class. So,
earlier entered additional method QSerialPortPrivate::setError() is
removed.
This patch with using of slots is a strategic step because in the future
in case of fix of a bug QTBUG-34946 instead of the QWinEventNotifier it
is necessary to use the QWinOverlappedIoNotifier which provides only
the signal/slot API.
Note: Also this patch implicitly fix a bug QTBUG-33938.
Task-number: QTBUG-33938
Change-Id: Iba9137e13e0bd6001c665434698b8cd965bc36e5
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
|\
| |
| |
| |
| |
| |
| |
| | |
Conflicts:
src/serialport/qserialport.cpp
src/serialport/qserialportinfo_unix.cpp
Change-Id: I12658a9f92c91ab83f404c93a311ef947d1f2503
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Do not try to work around the gcc/clang limitations with external variables
since some people have some concern that it may break the binary compatibility
due to the underlying representation layout.
Talk to upstream (gcc) if a warning for this is desired. It is out of scope for
us. C++14 will help with the [[deprecated]] attribute, but that is light years
ahead.
Task-number: QTBUG-35354
Change-Id: I1c4e0fb950469adceca85147084fa896503de78b
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
|
| |
| |
| |
| |
| |
| |
| | |
Task-number: QTBUG-35215
Change-Id: Id8175af00c431cc8f80f05d518b3a2ee0e8ac71b
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| | |
Task-number: QTBUG-35215
Change-Id: I53af65ba485216a6fde238245580fe0422a2ed9a
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| | |
Task-number: QTBUG-35184
Change-Id: Ie822943bcaf071024bc4f9782441b0534fcbd43d
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Unfortunately, 0973537c5471288726b929250199a41435729bee caused a serious
regression for obtaining detecting PCI serial ports without udev. This was
working before since the hard coded fallback mechanism has been up to this task,
but the sysfs alternative took the precedence over.
A note has also been added to the udev part of the algorithm because we will
need to obtain the vendor and product identifiers in a separate block in the
future. The vid/pid will be fixed in a separate change once that is well-tested.
This patch has no any risk as far as I can tell since it is just adding a new
short block to the existing code without much implication.
I have tested the change on Linux for my use cases with Qt 4 and then 5, and I
cannot find any regression. Everything works as expected both with udev and
without, i.e. with sysfs.
Thanks go to Aaron Linville for reporting the issue.
Task-number: QTBUG-35064
Change-Id: Ieae6f3e94559fb61ede67ff1d8810bb686d46383
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
Reviewed-by: Aaron Linville <aaron@linville.org>
|
| |
| |
| |
| |
| |
| | |
Change-Id: I58ee8bf1913c773f345b23bd01d1cc3fefc5f6ac
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|
| |
| |
| |
| |
| |
| | |
Change-Id: I211a417d1e4af8e22a2fd736093f099b8fd15cd8
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|
| |
| |
| |
| |
| |
| | |
Change-Id: I7e12e26e6676ecb064eb06599234d2844041335d
Reviewed-by: Denis Shienkov <denis.shienkov@gmail.com>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The current "Setup Class GUID" approach ignores the devices which has
been created by the drivers with a custom setup GUID's (e.g. Eltima
Virtual Serial Port Driver and so on).
In this case need to add each new custom GUID to the list. But it is
not a good solution and shall be used as a last resort.
The good solution is use of the "Device Interface Class GUID" instead
of the "Setup Class GUID" because it is the recommended way:
http://msdn.microsoft.com/en-us/library/windows/hardware/ff545036%28v=vs.85%29.aspx
http://msdn.microsoft.com/en-us/library/windows/hardware/ff545046%28v=vs.85%29.aspx
A new approach uses the standard GUID_DEVINTERFACE_COMPORT and
GUID_DEVINTERFACE_MODEM GUID's which shall be used for the development
of any serial port drivers. Besides requires adding of the additional
DIGCF_DEVICEINTERFACE flag for the SetupDiGetClassDevs() function,
see MSDN:
http://msdn.microsoft.com/en-us/library/windows/hardware/ff551069%28v=vs.85%29.aspx
It should be noted that this approach also can ignore some devices
with drivers which don't use the recommendation. Earlier already
there were attempts to implement this approach but unsuccessful,
see more 3bfe998860e0bfd3ce48784188eaa0d58bf86da1 and
bcdeb589dd9d3b0ff7270d380d043b1882eefdc0 commits.
Thus, it seems effective to use both "Setup Class GUID" and
"Device Interface Class GUID" approaches that will complement each
other. In this case can be keep only the standard GUID's.
1. Now all GUID classes are in the QUuid form and stored in the
constant list of pairs as QPair<uuid, flags>. It allows to avoid the
old heritage related to the pure-C coding style.
2. The second "flags" parameter from the pair is the DWORD input
parameter for the SetupDiGetClassDevs() function which specify the
control options that filter the device information elements. It
allows to pass the appropriate flags for the each used approach
independently.
Tested with the cenumerator example on Windows with Qt4 and then Qt5,
no regressions found.
Task-number: QTBUG-33313
Change-Id: I478905e8ae4b79b7d967e0727d2e1620ad667177
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Laszlo Papp <lpapp@kde.org>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The algorithm before the commit eca1a6d30a98861cb811cb2faf8d01334e60dd7c
could process the write events from the several notifiers, which could be
created in data transfer operation.
But now the write notifier always is in one instance, so can do following
optimizations:
* The method lookupFreeWriteCompletionNotifier() now is renamed to the
lookupWriteCompletionNotifier() and returns the pointer to the existing
notifier without creating them (by analogy with others lookupXX() methods).
* Now there is no need to control the write notifier state (to do enable
or disable) in data transfer process. The current algorithm doesn't require
it and the write notifier enabled once when port opens.
* Use the open mode flag instead of low-level events mask to create the
write notifiers. Also, this handling was added similarly to the read
notifier.
Tested the work with the terminal, the creader(a)sync, the cwriter(a)sync
examples with Qt4 and then Qt5 on Windows 7/8 with use:
* the virtual serial ports (the com0com project)
* the physical on-board serial ports and the USB serial ports
Change-Id: I8bc6a025a00dde986fde38c052d3ade4215938ec
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
Reviewed-by: Laszlo Papp <lpapp@kde.org>
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
For more detail, see remarks from the MSDN:
http://msdn.microsoft.com/en-us/library/windows/desktop/ms684342%28v=vs.85%29.aspx
Change-Id: Id6e5adc48e8a4648f1aabba598696b34c658a159
Reviewed-by: Laszlo Papp <lpapp@kde.org>
Reviewed-by: Sergey Belyashov <Sergey.Belyashov@gmail.com>
|