These changes both simplify the Java code and improve performance across the
JNI boundary.
This also fixes the AnalogCrossConnectTest by adding delays to
setInterruptHigh() and setInterruptLow() to ensure the change in voltage has
time to propagate and extends the timeouts in AbstractInterruptTest.
Detailed changes:
Hoisted status checks to C. This avoids the need to create direct byte
buffers (expensive) and significantly simplifies the Java code. The C code
now directly generates the exception or reports the error to the DS.
The JVM pointer is now a global across the JNI, initialized by the OnLoad
function, avoiding the need for some of the class-specific initializers to
get this pointer for callbacks.
Opaque pointers (such as ports) are now passed as long values rather than
with a ByteBuffer wrapper.
Added extern "C" to source files. This allows earlier detection of JNI
definition mismatches to the Java source headers.
Changed JNI signatures to more closely match HAL signatures (in particular,
boolean is now universally used instead of byte for HAL bool, which cleans
up mapping back and forth to 1/0 from true/false).
Change-Id: I4ea0032cabb0871cd74106a3a70d947258c29d2d
The HAL functions often call NI functions that may expect status to be
initialized on entry. Java does not guarantee direct pointer allocation
memory to be initialized, although that may have been occurring by accident.
Change-Id: I5e3a553f3a7be8de3716ccfc13e6ca1cb4f2a552
The bindings only wrap the HAL interface, rather than the entire C++ Notifier,
as I ran into issues trying to wrap the whole Notifier (all the existing
bindings only wrap HAL components, so wrapping stuff in :wpilibc is
unexplored). As such, the new edu.wpi.first.wpilibj.Notifier is just a
re-implementation of the wpilibc/.../Notifier.cpp.
The purpose of doing this bindings is to allow Java users a better option
for running tasks which require good timing (such as control loops). The
previous method used java.util.Timer to schedule a task, causing various
issues. Although this update does improve things, Java loop timing is still
substantially worse than that of C++, and, even worse, if Java decides to call
the garbage collector at the wrong time then the loop can be delayed by
multiple milliseconds and the next iteration will be shorter to account for it
(although this particular behavior could be updated).
A few notes on individual components:
-the HAL Task.hpp and Task.cpp were modified due to compilation/linkage
issues with the JNI bindings. Nothing substantive changed.
-NotifierJNI was added to the build files for gradle.
-HALUtil was modified to include a function for getting the length of a C
pointer, rather than relying on it being 32-bit.
Change-Id: I966512d8a82c2a438ed8c8bbcc6cdc6ed186d0f2
