Allow to inject keypress events to remote peers (most notably Android devices)
Notes / open issues / possible improvements:
- For the json-payload I used the syntax of the key-events as sent by mousepad-plugin with the addition of a "sendAck"-flag. If "sendAck" is set to true the remote peer should echo a key-event if it could be handled, thus allowing the local client to find out whether the key was accepted. For performance reasons, it's allowed to send multi-char strings in the "key" property (performs much better if you send a whole string via "echo '...' | kdeconnect-cli ..." e.g.)
- kdeconnect-cli: For now takes a string and transforms it into single key-events for visible characters only. In a first implementation I used a kbhit() helper that used termios.h to catch and relay keypresses interactively (including some special-events), which was not optimal. A better approch might be to use linux input-api directly. Would this be an option regarding cross-platform compatibility or can I assume to develop for Linux only? Being a command-line guy, I'd really like to have a fully featured kdeconnect-cli interface ;-)
- Factor out the Qt::Key-to-internal keymap to some core-helper because it corresponds to the mapping in the mousepad-plugin?
- The plasmoid is not perfect as it is: A single line containing a non-echoing TextField (i.e. it eats all the KeyPress events), and only ack-ed keypress-packets from the peer device are injected if they contain visible keys. Advantage: the user sees whether his key-presses are accepted by the peer device. Disadvantage: The echoed text does not correspond 1:1 to what is shown on the peer's display, user might be confused when typing without success. I played around with different variations each of which with its proper shortcomings:
1. An echoing Textfield for typing: Has the advantage that the user can directly see what he is typing, which makes interaction in the typing field easier, BUT messes up interaction if the Editor on the peer is changed silently and does not notify the user if his keypresses are not handled by the peer.
2. A non-echoing TextField for typing PLUS a readonly one for printing visible echoed keys. Disadvantage: same as for the previous one and uses more space on the plasmoid.
Comments? Ideas?
REVIEW: 129727
BUG: 370919
Therefore notify the client that we're deleting all notifications.
Otherwise the client wouldn't notice this and the notifications would still
be available in the plasmoid.
REVIEW: 127582
The server supports this device state (e.g. if a device was saved as paired,
but isn't reachable), so let's expose it to the model.
The display filter should take care of excluding such devices if
nessecarry.
To make sure that such devices aren't shown in the kcm (we can't do anything
useful with them at the moment), set the display filter accordingly.
REVIEW: 127553
There were two clearNotifications() calls when refreshing the notification
list. As there happens nothing relevant in between them, the second call is
useless.
REVIEW: 127542
Listen to dbus registrations. Remove all devices if the service dies,
create all devices back when restored.
Same for the NotificationsModel.
REVIEW: 124183
Removes the id->row mapping. It was broken as the indices weren't updated
when a row was added or removed.
Stop exposing Device::id through dbus, just store it in the interface
class.
* It's redundant as it's part of the interface definition
* If we keep it it will save us quite some dbus round-trips
* It will be reliable, as nowadays id() sometimes QString() if the dbus
connection is invalid.
REVIEW: 124158
Add API to call random public plugin methods
qdbusxml2cpp is not generating the Q_PROPERTY attribute. For now [1], add
it ourselves by addding a proxy signal.
REVIEW: 124077
Uses KPluginLoader+KPluginMetadata instead.
Describes plugins in json instead of desktop files. These desktop files are
then embedded into the .so file. All the plugins will be in a kdeconnect/
directory, and those will be the ones to look for.
Note it doesn't drop the KService dependency as KIOCore is a KDE Connect
dependency as well.
REVIEW: 123042
Make the population asynchronous on dbus. Wait for the device list and
don't populate it until it's ready. There's no reason to block.
Don't emit after endInsertRows. It doesn't help, as the views will have
already refreshed.
Benefits:
- We had config files in different paths, now they will be centralized.
- Daemon, LanLinkProvider and NetworkPackage depend only on Qt now.
- KConfig is accessed with strings as keys, where a typo can go unnoticed.
- Daemon is now thinner, as it doesn't need to initialize config values.
- QCA::Initializer is in a single place now (was in Daemon and Plugin).
Straight forward port from QJson to the native Json support.
As a note I have added 2 helper functions that make the code bit more
readable (object2qvariant and qvariatn2qobject).
Ported using KDELibs4Support, for a smaller delta, so we can keep
developing on master until we decide not to.
At the moment, it builds and installs but tests don't pass because
of a QCA2 initialization problem I didn't manage to debug yet.
CCMAIL: kdeconnect@kde.org
