MjpegServer uses the timeout to generate keep-alives to any clients
(which helps detect disconnects and avoid stale client threads).
CvSink GrabFrame now defaults to a timeout, but the timeout can be
changed by the user, or the old no-timeout version is now available
as GrabFrameNoTimeout.
Would throw if the camera was disconnected. We handle this properly at
the JNI level to not have this crash the entire program, but the error
is still kind of annoying, and not really an error.
Unfortunately, due to the way NT synchronization is currently performed,
this has unexpected and undesirable behavior: when a dashboard (or any other
NT client) is left running between code restarts, when it reconnects, any
code settings will be overwritten by the NT synchronization process. As
fixing this will require a fairly major NT change (and likely a user-visible
one), it's not desirable to do at this point in the year.
Instead, disable NT driven settings entirely (e.g. make the NT interface
publish only). To emphasize the read-only nature of the NT values, attempts
to change the NT values will be immediately overridden by CameraServer.
To better inform users about the actual property names (e.g. for use in their
code), the "raw_" settings no longer have "raw_" removed from their names
(they are still placed in the "RawProperty" subtable).
This change also contains a couple of Java fixes:
* Make getInstance() thread-safe
* Properly synchronize access to m_tables between multiple threads.
* Use Hashtable instead of HashMap.
When the page is loaded, if properties can be found they will
automatically be created on screen. They are currently not auto
updating. Raw values are currently disabled because of this.
Currently if using a separate compiler prefix, it would get published to
the arm classifier. This modifies so the output suffix can now be
specified (e.g. "hf" for armhf).
Currently if using a separate compiler prefix, it would still get published
to the arm classifier. This modifies so a classifier suffix can be used to
disambiguate arm from armhf.
* Java SendableChooser now decorates with type (non-breaking change)
* C++ SendableChooser now is templated on the type instead of using void* and stores values (breaking change)
* C++ SendableChooser now uses llvm::StringMap instead of std::map
