2020 port

[prensing]

I think this is ready, but this is my first PR to Robotpy, so I expect there are tweaks. A few notes:

• this does not include the new classes RawSource and RawCvSource (and Sinks). The code says they are incompatible with CvSource, etc. Not sure what people will want with them
• I test compiled against both OpenCV 3.4 and the just released 4.2.0. There was an small change in the API for UMat converter, but I #ifdef-ed it away.
• There is no code in setup.py to locate the OpenCV install. OpenCV4 got installed a bit oddly for me but I don't know if that is a new feature, so I worked around it on my system.

[prensing]

Oh, also, the file "cscore_src/cscore/src/main/native/cpp/jni/CameraServerJNI.cpp" was part of the compile, which seemed unnecessary. I updated setup.py to exclude it.

[auscompgeek]

I would prefer using AccessFlag = int;.

[prensing]

OK. My C++ is rusty, so that is new to me. Not sure it is relevant, but PyInt_Check uses #define just above that.

[auscompgeek]

OpenCV4 got installed a bit oddly for me

Could you explain?

[prensing]

OpenCV4 got installed a bit oddly for me

Could you explain?

I did a fairly simple install with the plain OpenCV 4.2.0 download. It was set to install into /usr/local. The include files installed into directory /usr/local/include/opencv4/opencv2. When the code tried to include "opencv2/xxx.h" it could not find it. I worked around it by linking /usr/local/include/opencv2 -> opencv4/opencv2.

I thought about adding code to setup.py to search for the OpenCV install, but a clean solution was not obvious.

[prensing]

The auto format checks appear to be flagging only code in cscore_src, so presumably should not be fixed?

[auscompgeek]

I would recommend actually reading my notes in #39. Plenty of notes about that there.

[auscompgeek]

The auto format checks appear to be flagging only code in cscore_src, so presumably should not be fixed?

Ha. Could you add cscore_src to the black exclude in pyproject.toml?

[prensing]

I will look through #39, but have to go do other stuff for now. Everything seems to be working with OpenCV4, but my testing was limited to trying our 2019 vision code (and not for long).

Do we want to maintain compatibility with OpenCV3? If so, the include path should be flexible...

[auscompgeek]

Do we want to maintain compatibility with OpenCV3?

Yes. We will probably still be using OpenCV 3 for the roboRIO for 2020, matching WPILib. I would also like to be able to continue building against distro packages.

[prensing]

I updated setup.py to use the pkgconfig package. It worked correctly for OpenCV4 (compiled localled) and the Ubuntu-provided OpenCV 3.2 packages. pkg-config is available for non-Linux (incl. Windows) so I coded it to use "pkgconfig" (the Py module) on any platform if it loads.

Remaining possible issues that I can think of:

• pkgconfig is not required by the build, but that is probably good
• install docs should be updated to mention the CMake "-D OPENCV_GENERATE_PKGCONFIG=ON" flag for OpenCV.

[prensing]

So I think the JNI filtering is done. What is the etiquette for Resolving conversations (assuming you want that)? May I, or is that left to the person who started it?

[virtuald]

I suppose whoever thinks it's resolved?

[auscompgeek]

Anything else we need to do here?

[prensing]

From my POV, I believe this is done.

[virtuald]

I haven't built it yet, but it looks fine and the tests pass (which includes a build) so I'm going to merge this.