This implements the proposal discussed in the re-opened #279 to add a
pin_factory argument at the device level and remove the ability to
specify a pin instance to device constructors (they now only accept a
pin specification).
Note: there's still a couple of bits to tidy up (tests on "real" Pis,
and pin_factory.release_all needs refinement) but the test suite is now
at least capable of passing on a PC.
Factor out hardware SPI pin numbers; this is a effectively a rebase (and
minor modification) of @lurch's original patch which conflicted after
the merge of the remote-spi branch.
Ensure SourceMixin descendents shut down the source prior to closing.
Furthermore, make sure devices are closed before pin factory shuts down,
and that pins have a strong reference to their owning factory (to
prevent losing the factory before the pins).
Some fairly major changes to ensure that the Pin.when_changed property
doesn't keep references to the objects owning the callbacks that are
assigned. This is vaguely tricky given that ordinary weakref's can't be
used with bound methods (which are ephemeral), so I've back-ported
weakref.WeakMethod from Py3.4.
This solves a whole pile of things like Button instances not
disappearing when they're deleted, and makes composite devices
containing Buttons much easier to construct as we don't need to worry
about partially constructed things not getting deleted.
Ensure LEDCollection cleans up upon construction failure, rename some
internals to be a bit more obvious, rename PinGPIOUnsupported to
PinUnsupported, and some other stuff I've forgotten!
Sorry! Dave's messing around with the pin implementations again.
Hopefully the last time. The pin_factory is now really a factory object
which can be asked to produce individual pins or pin-based interfaces
like SPI (which can be supported properly via pigpio).
