We overwrite the current formatter to display or not
the prompt depending if we are in the prompt or not.
To prevent races between the background events and our
current configuration, we acquire/release the lock on
the current handler.
Only list and listen commands are currently implemented.
The Ctrl-C handling has been a little bit tricky. The default GLib
mainloop tends to add its own SIGINT handler, which prevents us to
gracefully handle the KeyboardInterrupt exception during cmdloop().
So we need to create the mainloop in TuhiKeteShellWorker directly,
but bypassing the GLib.Mainloop() python facility.
For the various commands, it is easier if we have a common interface
than just a simple function call.
Each current command runs something before the mainloop is created, and
then something after the mainloop is terminated. Having such a worker
allows us to have only one place where we start the mainloop, meaning
that the interactive prompt will not try to start it more than once,
and most above, will not kill it in the middle of a command while other
commands are still running.
Again, very rudimentary but we're planning on making kete a bit better anyway.
Also note that this doesn't make use of the tuhi.drawing module on purpose,
parsing the format twice avoids some bugs and also keeps tuhi-kete separate
from the tuhi daemon.
Part of #7
As we're planning to cache the data locally, the timestamps are a
unique-enough way that makes it possible to access a specific drawing.
And this way we can also delete some drawings without all other indices
shifting around.
Fixes#16
If we're losing the bus name (i.e. we can't get it on startup) there really
isn't much we can do other than fail miserably. But in passing signals around
we can't do exceptions, so we have to move the mainloop to Tuhi so we can
quit() it on error.
Fixes#25
Basically copied from the device's Listening approach.
While it's possible to have multiple clients searching at the same time it's a
niche case and the effort at fixing the race conditions that come from that is
likely not worth the effort.
Let's add multiple simultaneous clients when we have a real need for it.
TuhDevice.paired is set on every device update (RSSI changes!) and that sends
a GObject.notify() for the property. Filter those, otherwise
we're just spamming dbus with PropertiesChanged notify events even
though nothing has actually changed.
We likely get multiple 'udpated' notifications as the RSSI property changes,
all causing a Connect() on the device and an ugly error message (that we used
to catch and print). Make that error message prettier.
When bluez restarts (or the tuhi daemon restarts), the values we have
in the bluez device's field ManufacturerData are quite not accurate.
When bluez restarts they are empty, and if the last time we saw the
device was for the pairing process, the device will still be marked
as in the pairing mode.
So we should mark the cold-plug sequence differently from the hot-plug
one, and we should be more confident in the current configuration we
have stored to export the currently known devices over dbus.
Fixes#13
If the sender disappear, we should stop listening for incoming events.
We match the Start/StopListening() that way, but if a client forgets
to call StopListening() before leaving and some data are being retrieved,
it's not our problem.
Benjamin Tissoires