-D and -I are preprocessor directives, they are therefore to appear
in CPPFLAGS, not CFLAGS. (It is unfortunate that pkg-config does not
make the distinction / or it was misnamed, because only -D/-I make
sense to be emitted by pkg-config in the first place — anything else
(-f/-m) has the potential to mess up someone's compilation. So
pkg-config's Cflags is actually used to convery Cppflags.)
When using automake, all AC_SUBSTed variables are made available as
make variables, so that they can also be set at make time. Therefore,
they need not use the @harcoded@ style.
When files are to be placed not in libexecdir but a subdirectory of
it, automake has a variable name reserved for exactly that purpose
(and a default value, which Makefile.am will override), called
pkglibexecdir. Let's use it.
This adds wrappers function to interface with GLIB mainloop so
applications can use mainloop functions no matter what is the underline
implementation.
Note: Most functions are not actually implemented on purpose since both
io and timeout functions already exists for GLIB covering the same
functionality.
When building from out of tree, and the top build dir was specified as
an absolute path, the linked headers in ${builddir}/lib/bluetooth were
broken. This patch fixes it by relying on make's abspath macro as
opposed to the path concatenation.
I tried to test all basic and most common use-case scenarios here and
some more weird as well. It tests cases where the parser needs to handle
problematic MIDI messages as well as ALSA Sequencer events.
It is really simple to add new tests, so others are welcome to do so.
This plugin implements the Central role of MIDI over Bluetooth
Low-Energy (BLE-MIDI) 1.0 specification as published by MMA in
November/2015.
It was implmemented as a bluetoothd plugin because of latency requirements
of MIDI. There are still room for improvements on this regard.
Like previsouly mentioned, it only implements the Central role, but
since all parsing and state-machine code is in libmidi.[hc] it should be
simple to implement the Peripheral role as a GATT service as well.
Files added:
* profiles/midi/midi.c: Actual GATT plugin
* profiles/midi/libmidi.[ch]: MIDI parsers
Techinal notes
==============
This plugin doesn't require any new threads. It relies on notifications
from a device to parse and render proper events that are queued in the
kernel, causing no blocks at all. Even if an error occur, it will be
handled and returned control to bluetoothd.
It also adds a new file descriptor to be read using struct io. That is
necessary to read events from applications and render raw BLE packets to
be sent to the device with a write without response command. It doesn't
block as well.
This patch introduces ALSA as dependency. But this feature is disabled
by default. To enable it, pass --enable-midi to the configure script.
Even though this introduces ALSA dependency, it is not an audio plugin.
It is rather a MIDI plugin, which is a byte stream protocol with low
throughput but requires low-latency.
Observations
============
I have tested on a normal laptop Arch-linux (x86_64) and a Raspberry Pi 2
(ARM Cortex-A8) and it works very well. As I mentioned, the latency can
always be improved.
I will still maintain a personal branch on my github[1] so others can
contribute there and I can test before sending to BlueZ.
IMPORTAT: the timestamp support is incomplete since ALSA doesn't support the
way MIDI over BLE expects (asign timestamp to an event without scheduling).
We are working on ALSA to support this.
Credits
=======
I would like to send kudos to ROLI Ltd. which allowed my to work
on this as part of my full-time job.
[1] https://github.com/ftonello/bluez/
