Permalink
Comparing changes
Choose two branches to see what’s changed or to start a new pull request.
If you need to, you can also or
learn more about diff comparisons.
Open a pull request
Create a new pull request by comparing changes across two branches. If you need to, you can also .
Learn more about diff comparisons here.
...
7
contributors
Commits on Jan 23, 2017
Commits on Feb 4, 2017
-
Use a smarter algorithm to get the available ports on Linux
We iterate over /sys/class/tty/* looking for tty's. We check that there's a device associated to ignore psuedotty's and that they must have a kernel major:minor number. For platform:8250 we open the device and check further. Only devices that exist in /dev, and are user accessible which appear real are shown (multiple nodes to the same device are ignored). The description returned is the name of the device given by udev in the uevent file. The libnserial library is updated to version 1.1.0 as we now have a new C-API. The SerialPortStream remains compatible with libnserial 1.0.x, and if that is installed on the system, SerialPortStream will revert to old behaviour. Tested on Linux. Simulated v1.0.0 library that fallback works (with EntryPointNotFoundException). Issue: DOTNET-34
-
Commits on Feb 22, 2017
-
These items have been in the repository for quite some time and are arguably no longer relevant. If it's required, then put them in a bug tracking system instead.
-
Support .NET Standard 1.5 (.NET Core 1.0 and .NET FX 4.6.x)
Original patch from Jaime Olivares <jaime_olivares@hotmail.com>, reworked as described in the following sections: Rebased .NET Core, enabling GetPort() functionality also for .NET Standard. .NET Standard doesn't support ApplicationException, so a more explicit InternalApplicationException is created. Functionality for getting the serial ports was reenabled. The libnserial v1.1 adds this functionality. Unfortunately, .NET standard doesn't support System.EntryPointNotFoundException (it's raised on Windows, but it's not exposed). So there's an ugly hack that in case of an exception when calling the method under Unix, we remap all exceptions to System.EntryPointNotFoundException. Note: On Windows, you'll no longer get port descriptions, as the WMI functionality is not available under .NET Standard. Note: On Linux, exceptions aren't mapped properly under .NET Standard as there's no mapping of errno to an appropriate exception, like there is for Mono. You won't be able to capture them properly. This is to be considered a bug. Note: {Begin|End}{Read|Write} isn't defined in .NET Standard. They use ReadAsync and WriteAsync instead. Issue: DOTNET-72, #10 -
Added proper exception support for Linux.
For .NET Standard and Mono, we now do our own mapping of 'errno' to constants defined by libnserial. This allows us to raise the correct exceptions in .NET Standard. Mono now raises more exceptions than previously (may cause a breaking change in existing code under Linux). This library will not work on .NET Standard 1.5 under Unix operating systems, unless libnserial.so.1.1 is in the path. The older library does not provide the mapping routines. On Mono, you can continue to use the old library and we use the old mapping algorithm. Remember, as soon as you update the library, you'll get the new exceptions automatically. Issue: DOTNET-75
Commits on Mar 25, 2017
-
Remove platform check on Linux for Mono.
A platform check was introduced to ensure that for NET Standard 1.5, the user has the library nserial 1.1.0 or later installed. This is useful for Mono, but not required. Issue: DOTNET-81
Commits on Apr 7, 2017
-
Close a serial port properly on Linux with Native Lib.
In some cases after closing the serial port and immediately trying to reopen it, we'd return EACCES indicating that the serial port is in use. This occurs because of the usage of the ioctl(TIOCEXCL). It's not possible at all anymore to open the serial port. So just before we now close the serial port, we reset the exclusive flag with ioctl(TIOCNXCL), which allows the serial port to always be reopened. Tested by modifying the Mono test case to run 100 times, which now passes (it didn't before). Issue: DOTNET-82
Commits on Apr 8, 2017
Commits on May 5, 2017
-
Abort pin monitoring on Linux for devices that don't support it
For some serial devices, such as ttyACM* USB devices, the ioctl(TIOCMIWAIT) returns an error. Now if an error occurs, we propogate it to the .NET implementation which already correctly ends the monitoring thread. Previously, the error was ignored in a loop resulting in 100% CPU utilization. Issue: DOTNET-85, #22
-
Fix race condition when checking/aborting modem events.
The method serial_waitformodemevent should only be called at most once. There is already some code to protect the structure used for waiting for the event, but none when cancelling, or when waiting for the event twice (meaning that the code was incomplete and did nothing). This commit finishes the implementation. In addition, we now report errors when creating the critical section, and errors when creating the cancellable thread. Issue: DOTNET-86, DOTNET-87
Commits on May 6, 2017
-
Make usage of pthread_setcancel{state|type} portable
According to the linux manpage, the second parameter may be NULL, but this is not standard in the Posix specification. We should provide a pointer to an int, even if we don't use it. Added test cases to ensure that the function serial_waitformodemevent can be aborted by serial_abortwaitformodemevent. Tested that this works. We tested also what happens if the type is PTHREAD_CANCEL_DEFERRED instead of PTHREAD_CANCEL_ASYNCHRONOUS and it turns out that the call to ioctl(TIOCMIWAIT, ...) is not a cancellation point. Issue: DOTNET-93
-
Properly handle errors from pthreads
Pthreads return the error number, not -1 like just about every other system call. So handle this properly. Issue: DOTNET-94
-
We have to add delays between setting and checking CTS/RTS and DTR/DTS
Implement a 10ms delay in the test case. It appears that for the USB serial device used for testing, setting the RTS/DTR pins isn't done by the OS immediately. Issue: DOTNET-96
-
Correct race condition when aborting a modem event
There was the possibility of returning an error when aborting a modem event, depending on timing if a signal arrived at the same time. There should be no error from aborting a modem event even there's nothing to abort. Issue: DOTNET-95
-
-
Treat EINTR as a non-fatal interrupt
If a signal occurs for whatever reason, the OS may return EINTR as an error code. This was being treated as a fatal error in the .NET implementation. libnserial now returns zero for serial_read and serial_write as is the case for EAGAIN / EWOULDBLOCK. The .NET code was optimized slightly so that in case zero is returned, it skips a buffer check and tries again (this change is non-functional). Issue: DOTNET-91
-
Manual tests that are explicit won't be automatically run from NUnit in the VSIDE, MonoDevelop IDE or the NUnit runner unless explicitly set. Issue: DOTNET-88
-
Thread names on Linux too long
Shorten the names of the .NET threads (I/O and pin monitoring) to something shorter. Issue: DOTNET-92, #24
Commits on May 7, 2017
-
Provide an implementation for DiscardInBuffer and DiscardOutBuffer
The implementations on Linux remained not implemented, throwing an exception. The issue does not affect Windows. Issue: DOTNET-97, #26
Commits on May 8, 2017
-
Use a SafeHandle for libnserial
Previously used an IntPtr. Wrapping this around a SafeHandle implementation protects against some memory errors. Further, the Dispose pattern can be simplified and we don't need a Finalizer (which we were previously missing). Issue: DOTNET-100
-
Don't execute some test cases under the Monodevelop IDE
The test cases run and cause follow up failures, making testing tedious. Add value to test cases by not running those we know cause issues. Issue: DOTNET-101
Commits on May 13, 2017
Commits on May 22, 2017
-
Support devices that don't work with overlapped WaitCommEvent.
Some devices, notably a Arduino Uno clone (with CH340 on board), would return error 87 (ERROR_INVALID_PARAMETER) on a call to WaitCommEvent, but it does work with the GetReceiveStats (Windows API ClearCommError function) to know when data is available to read. The flush functionality should still work on write, even without the TX_EMPTY event that we'd miss with WaitCommEvent, because we set this every time our internal buffer is empty. All unit test cases on Windows with the COM-0-COM driver work. Testing on the Arduino CH340 borad done by GitHub user AndreiGorlov. Issue: DOTNET-98, #25
Commits on May 25, 2017
Commits on May 26, 2017
Commits on Jun 16, 2017
-
Updgrade project to work with .NET Core 1.0.4
Originally provided as a pull request from Danny <dannyb2100@hotmail.com> in #28. Moved the csproj for netstandard 1.5 in the same folder as the rest, provided a solution file for VS2017. The dependencies in the CSPROJ file are now 4.3.0 to match that in the NuSpec file for .NET STandard 1.5. Build script and NuSpec was updated for the new binaries. Issue: DOTNET-104, #20
Commits on Oct 31, 2017
-
Correct paths for CMake in README.md
Issue found by user ardave with #40. Additionally corrected a similar sentence above to make the instructions consistent.
Commits on Apr 25, 2018
-
Increase timeout for test case ListPorts
When running on a Windows machine that has 4 emulated ports and four bluetooth ports, the management engine of the desktop version of windows may take some time. 10 seconds is not enough. The code that is taking its time is: public PortDescription[] GetPortDescriptions() { ... ManagementObjectCollection objects; using (ManagementObjectSearcher q = new ManagementObjectSearcher("select * from Win32_SerialPort")) { objects = q.Get(); using (ManagementObjectCollection.ManagementObjectEnumerator enumerator = objects.GetEnumerator()) { while (enumerator.MoveNext()) { ManagementObject current = (ManagementObject)enumerator.Current; string k = current["DeviceID"].ToString(); if (list.ContainsKey(k)) { list[k].Description = current["Name"].ToString(); } } } } ... Issue: DOTNET-152 -
SerialPortStream: The IsDisposable flag should not be publicly settable
Additionally, clean up Dispose() to reduce indenting, and properly do a GC.SuppressFinalize. The GC change is to meet the Dispose programming pattern and is required in case the classes are derived from. That these classes aren't derived from means that this change is cosmetic and a coding standard change only. All unit tests pass on Windows. Issue: DOTNET-154
-
Use constant for infinite timeout
Code uses -1 in many places to mean an infinite timeout. Instead, use System.Threading.Timeout.Infinite to differentiate between other uses of -1. This code change should not result in a change of the binaries, that constants are not stored in code. Issue: DOTNET-155
-
Avoid deadlock in event handling
When an event occurs, a "m_EventLock" was taken to check the number of bytes received which took the lock "m_ReadLock". Separately, the main I/O thread would take the lock "m_ReadLock" and may indicate that an error occurred trying to take "m_EventLock". This may result in a deadlock situation of a read event occurs simultaneously with an error event. Issue: DOTNET-156, #50
Commits on Apr 26, 2018
Commits on May 13, 2018
-
Use correct header when parsing for serial ports on Linux
When compiling on GCC 7.3.0 on Ubuntu 18.04, we got the error: .../dll/serialunix/libnserial/portlinux.c:243:13: warning: In the GNU C Library, "major" is defined by <sys/sysmacros.h>. For historical compatibility, it is currently defined by <sys/types.h> as well, but we plan to remove this soon. To use "major", include <sys/sysmacros.h> directly. If you did not intend to use a system-defined macro "major", you should undefine it after including <sys/types.h>. int major = major(sb.st_rdev); ^~~~~~~~~~~~~~~~~~~~~~~~~~ Include the correct headers and the warning goes away. Issue: DOTNET-158 -
Test Cases: Fix compatibility issues with GCC 7.3.0
When compiling on Ubuntu 18.04, the unit test cases fail to compile due to two changes: * 'auto_ptr' is considered deprecated, and unique_ptr should be used instead. * char is signed, and values greater than 128 raised an error due to narrowing Issue: DOTNET-159
-
Bump version of libnserial to 1.1.3
Ran all unit tests and pass where expected. Issue: DOTNET-160
Commits on May 29, 2018
-
-
Sign the .NET Standard 1.5 assembly
Introduced 'Signed_Release' build configuration to netstandard 1.5 project; enable signing for that configuration. Adapted build script to use Signed_Release build configuration for dotnet projects Issue: DOTNET-161, #55
Commits on May 31, 2018
-
Reduce number of objects created when Writing, Reading, Flushing
Instead of creating a new array of WaitHandle objects on everytime a read, write or flush is done, prepare the arrays in the constructor and use that. Looking into the .NET 4.7.1 reference sources, the method WaitHandles.WaitAny makes a copy of that array too, so there is no need to create it twice and burden GC0. Issue: DOTNET-162
-
Reduce CPU and Memory if tracing is disabled
Before printing out a trace, first check if it would be traced with the TraceSource.ShouldTrace command. Particularly, to make this as fast as possible and have the possibility for inlining, an internal implementation is made caching what trace events may be traced in a bit field. The GC is improved as we no longer need to create arrays for the TraceSource.TraceEvent(xxx, params object[] args), where the compiler will allocate an array automatically, and then box any value types into the array thus creating more objects on the GC heap. Issue: DOTNET-163
-
Reduce number of WaitHandles created in I/O loop
Instead of creating a new array of WaitHandle objects on every iteration of the I/O loop, we keep a fixed set of arrays and reuse them. Issue: DOTNET-164
Commits on Jun 1, 2018
-
The Enum.HasFlag operation results in two boxing operations and is significantly slower than using integer boolean arithmetic. The boxing arises through converting the Enum type to a reference type, so that HasFlag can be called. The second boxing operation is converting the flag to test from an enum type to a reference type for the test. This change would also slightly reduce CPU load. Issue: DOTNET-165
Commits on Aug 17, 2018
-
Update BUILD.md to be Markdown compliant
Make minor formatting to the BUILD.md document
-
Fix minor warnings indicated by VSCode
After installing VScode and installing the suggested C/C++-plugin, VScode reported some warnings. Although these don't affect the code, they might improve readability.
There are no files selected for viewing