As the notifierMutex is unlocked while the callback is being called in
alarmCallback(), it's possible for next to be invalidated due to a
cleanNotifier() call. Now, instead of deleting the notifier immediately,
add it to a freelist to be cleaned up at the tail end of alarmCallback.
Rename the following folders:
hal/lib/Athena -> hal/lib/athena
hal/lib/Desktop -> hal/lib/sim
hal/lib/Shared -> hal/lib/shared
wpilibc/Athena -> wpilibc/athena
wpilibc/simulation -> wpilibc/sim
Windows users may need to run gradlew clean after updating.
First part of the HAL changes. Returns -1 to 1 for joysticks instead of
-128 to 127. Implemented by replacing the old structure with a new
structure that uses floats instead of shorts.
Currently, about 5ms of every 20ms loop the DS thread would hold
the mutex, while grabbing data. During this time, and call to grab
joystick data would be blocked. This change grabs the joystick data
to a cache, and then grabs the mutex and moves the data references
around. This is much more efficient then the old code, and gives
teams more of their teleop loop time back for use.
Another major change this does is use preallocated arrays when entering
the JNI. Previously every JNI DS call would allocate an entire new array.
With a GC'd language where those arrays go on the heap, thats a problem,
and creates tons of garbage. That garbage is no longer created anymore,
as all arrays and byte buffers sent to JNI in the DS are preallocated.
In addition, GetJoystickName was always returning joystick 0 data, which
this fixes.
The PCM tests recently were failing by being slightly higher than voltage than expected. raised the tolerance to account for this.
Voltage was still clearly less than the on state while in the off state.
Change-Id: I8ac28867024f3ce7b4104fc03622e6133ac2b80f
{@inheritDoc} doesn't exist in Doxygen, so I removed instances of it from the C++ source. It isn't needed anyway because Doxygen automatically fills in base class documentation for member functions in derived classes if no documentation is given. The extension mapping for .inc files was accidentally removed during the build system changes so this commit adds it back in.
Change-Id: I599e3bdafd76ee7d78a66090d81707cda5a399d7
There was a lot of duplicated code between all 7 PWM speed controllers.
This moves all the duplicated code down to a base class, that will make
it easier to add speed controllers in future years if needed again. Also
if we need to add a method to all speed controllers, we only have to do
it in 1 place, and not 7.
Change-Id: I25eb1d097c0f5f7dbd7656db2f4a30d006d50f98
This actually isn't a large issue, since the HAL automatically creates a
notifier at the beginning to handle the time rollover, but during
testing without that, I noticed that deleting the last notifier would
disable notifiers for the entire lifetime of the program. This seemed to
be caused by the CLOSEST_TRIGGER static variable not being reset, and
causing new notifiers to never be added to the queue. This fixes that.
Not a bug that can be seen normally, but still a bug.
Change-Id: I6ac61c632918c90972cd0664aa833e3bc9e8c535
Ensure the POR, PWR, and CST error bits are not set when reading the sample.
Reported by: Jared Russell, Austin Schuh
Change-Id: Ib38178d6b51880480aee86e13c16f9ba89a80693
Java was using integer division for calulating Timer values, which
was causing Timer values to be concatinated, and only accurate
to the nearest millisecond. This changes it to use double division,
which matches C++ and makes timing much more accurate
Change-Id: I61b893f4ad5791bf4dfdd767ee4346c1ed4ea418
In the Java implementation, "result" is an instance variable (to avoid
needing to perform a memory allocation on each call to getAngle). However,
this means that calls to getAngle() from multiple threads can step on each
other. Synchronize the method to avoid this issue.
Reported by: Jackson Gray (http://www.chiefdelphi.com/forums/showthread.php?threadid=145124)
Note this does not synchronize accesses between configuration methods such
as calibrate() and accessor methods such as getAngle().
Change-Id: I9ecc962d4fba66f6134fd801cfd607f155b41fe5
The request is not guaranteed to be in a single packet, so it may be necessary
to perform multiple reads to get the complete request. The previous code
could unpredictably fail as it only performed a single read.
Fixes artf4817.
Change-Id: I7734a12e1a2542f5d7ca0889453c387f0bb30538
For Java, this is as simple as just adding it, as all motors already
have an implementation from MotorSafety that is correctly resolved. For
C++, I had to override StopMotor in the classes that descend from
SafePWM and explicitly call the SafePWM version. RobotDrive now calls
StopMotor on each of its SpeedControllers, instead of calling Disable or
setting the motor to 0.0 as it was doing previously.
Additional small formatting corrections to the previous commit starting
this were added.
Change-Id: Ie94565394927a910ce74bc628670ac3d658d8df9
also retuned the deviation over time test to make it pass more reliably.
Ran sucessfully 40/40 times run including several run with the entire wpilibj
Change-Id: I2e42c368fdb81f9206e15ce39878ea105da1e405
Fixes artf4809.
In both C++ and Java, error messages are being sent to both the console
(via stdout/stderr) and being reported via the HAL. We don't want to
remove the stdout/stderr reporting at present because users debugging only
via netconsole (e.g. using riolog) won't see the HAL-reported errors. Until
there's an alternative means to getting the HAL-reported errors to those
users, instead store the previous 5 error messages and don't duplicate them
more often than once per second.
Changes the error reporting from setErrorData() to sendError(), which
improves driver station error reporting. The "location" in C++ is given as
the immediate caller (e.g. the WPILib function). The "location" in Java is
given as the first user function in the call stack. Note the full call stack
is provided in both instances.
Change-Id: I590dd63dcb66825301ebb260aff00cd8d7d501ed
