With this change, “Reload Input” will discover newly attached devices
on macOS, which previously did not work.
Also add correct detection of disconnected devices, which previously
was not handled well, especially with the new ‘keep trying to open
devices’ behaviour of FGJoystickInput.
Should fix:
https://sourceforge.net/p/flightgear/codetickets/2259/
According to Pfeffer's report[1] and many other messages on the
Internet, joyGetDevCaps() can return the (so far) undocumented value
165, which some say is JOYERR_PARMS. Since this is a valid return code
for joyGetPosEx() (not joyGetDevCaps()!), handle a few return codes from
that function too, and don't throw an exception when joyGetDevCaps()'s
return value is not one of the documented ones (we simply can't trust
its documentation).
[1] https://sourceforge.net/p/flightgear/mailman/message/36657149/
Since asserts can be compiled as no-ops, throw an exception rather than
doing assert(false), in case the code passed to
joyGetDevCaps_errorString() is not one of the expected values.
Every code path that causes a joystick to report notWorking() should now
have an associated error message logged with class 'input' at level
'debug' or higher.
If joyGetNumDevs() > INT_MAX, casting it to an 'int' leads to undefined
behavior. On the other hand, after the 'ident >= 0' test succeeded, it
is perfectly safe to use static_cast<unsigned int>(ident), since 'ident'
is of type 'int'. Then we get to compare two unsigned ints, which is
well defined.
So far, the Linux PLIB implementation used to nuke any deadband set at
kernel level with jscal(1) as soon as the joystick was opened. This
broke joystick calibration for FG and unrelated programs, for people who
cared to tune their kernel-level deadband setting.
This fixes PLIB bug #47[1]. Note that the fix has already been in Debian
buster's plib for some time (plib 1.8.5-8 uploaded on 24 Jul 2017[2]).
[1] https://sourceforge.net/p/plib/bugs/47/
[2] https://tracker.debian.org/news/858564/accepted-plib-185-8-source-into-unstable/
