The protocol currently supports a ei_device.done event to notify
the ei client that the initial description of the device is complete.
For some future use-cases the client may need to futher negotiate
properties of the device. For example for tablet tools the client may
narrow down capabilities of the tool.
The sequence with the new request is thus e.g.
-> ei_seat.device
-> ei_device.name
-> ei_device.interface
-> ei_device.interface
-> ei_device.done
<- ei_device.ready
-> ei_device.resumed
In libei the request is sent automatically on unref of
the DEVICE_ADDED event. This makes clients immediately compatible
and for the typical (future) use-case of device configuration. Said
configuration will likely be handled in response to the DEVICE_ADDED
event anyway.
In libeis, a new EIS_EVENT_DEVICE_READY event that is sent when the client
sends that same event on the protocol, informing the EIS implementation
that this device is ready. For clients that do not support that version
the event is emulated immediately after sending ei_device.done.
This requires a flag bit to be long-term maintainable. The typical
EIS implementation currently calls eis_device_add() immediately
followed by eis_device_resume(). This doesn't leave any room to
wait for the client's ei_device.ready request.
One backwards-compatible solution could be to buffer the
eis_device_resume() until the ei_device.ready has been received but this
is fraught with hairy corner cases, e.g. if the client is a receiver
context we would also have to buffer all events immediately sent to the
client.
So instead, we have a flag in the context and if set by the caller, we
change the internal behavior to match ei_device interface version 3.
Part-of: <https://gitlab.freedesktop.org/libinput/libei/-/merge_requests/346>
This makes peck_new_context() take variable arguments in the style
key, v1, v2... where each value is defined by the free-form key.
What we need right now is the mode of the context, so let's add that.
In the future we will add more configurable bits here.
Part-of: <https://gitlab.freedesktop.org/libinput/libei/-/merge_requests/351>
If a receiver client stops calling ei_dispatch for a while eventually
we fill up the send buffer, causing messages to be quietly dropped.
When the client resumes the message stream resumes with whatever we send
next but that can leave the client in an inconsistent state.
deskflow hit this in the server-side where our event sequence of pointer
motion+frames eventually filled up the buffer, causing
eis_device_stop_emulating() to be silently dropped. On the next
InputCapture sequence eis_device_start_emulating() was sent to an
already emulating client (as seen by the client).
This patch adds a secondary queue, if we fail to send a message with
EAGAIN queue it up and flush that queue whenever the next message is
sent. Meanwhile any newly added messages go straight into that queue.
The caveat here: a nonresponding client will eventually trigger OOM,
there is no upper limit on the messages yet
This is the libeis version of
commit 69e973e6b3 ("ei: queue unsent messages for later delivery if our buffer is full")
Closes: #79
Part-of: <https://gitlab.freedesktop.org/libinput/libei/-/merge_requests/331>
This event is required to fix an issue with the current
ei_callback.done handling in libeis: previously we would imediately send
the ei_callback.done upon receiving ei_connection.sync from the client.
This results in incorrect behavior as we may have events in the queue
(and/or pending a frame) that the EIS implementation hasn't seen yet.
For example a client sending:
- ei_pointer.motion
- ei_connection.sync
- ei_device.frame
- ei_connection.sync
Will queue a motion + frame event on the EIS side during eis_dispatch()
but immediately receive done events for both sync requests.
This could be handled purely internally by keeping the sync event in the
queue but hidden to the caller - and automatically calling done when
it's that events turn, i.e. something like:
```
struct eis_event *eis_get_event(struct eis) {
struct eis_event *e = first_event(eis);
if (e == EIS_EVENT_SYNC) {
eis_callback_send_done(e);
eis_event_unref(e);
e = next_event(eis);
}
return e;
}
```
but that opens us up to a set of potential bugs detailed in
https://gitlab.freedesktop.org/libinput/libei/-/issues/71#note_2694603
So let's go the easy route by having a new event type that does nothing
other than eis_event_unref() in the EIS implementation. This way we can
queue the event and have everything behave in-order and as expected.
Closes#71
Part-of: <https://gitlab.freedesktop.org/libinput/libei/-/merge_requests/316>
We don't want to paper over bugs in the implementation, so let's make
any ei/eis error message with a bug in it fatal by default.
This needs to be disabled where we test for known-buggy client/EIS
behavior.
Leftover from 479bda259a (and possibly
others). This dates back to when a client could have restrictions
configured on the same fd. This is now all out-of-band (portals!) so the
compositor knows what the client is allowed to set up anyway.
No need for this (read-only) API here.
With the planned switch to a protocol supporting multiple interfaces
(a la wayland), a single version number is no longer useful. Remove this
API, we can add something more specific later if we need to.
This makes it easier to correlate a particular input transaction
(whether there are events or not) with out-of-band information like the
planned portal InputCapture::Activated signal's "activation-id".
libreis was intended for an intermediary to set some information that
the libei client cannot be entrusted with. In particular this was the
application name, the allowed capabilities, and some properties that -
once set - the client could no longer change (appid as probably the only
really useful one). The price for this was a rather complicated version
negotiation dance before the initial CONNECT request.
Now that we have a clear view of what's going to happen -
RemoteDesktop.ConnectToEIS and the InputCapture portal - there is no
longer any need for libreis. The extra information that libreis would've
sent is communicated out-of-band in both portals and are known to the
compositor at the time the connection is being established.
So we can simply drop this, it's no longer required and dropping it
makes the protocol significantly simpler anyway.
Let the client set the version number it wants on Connect. There is new
public API to query the client/server's version as set once the connect
finished (eis_client_get_version() and ei_get_version()) but there is
currently no public API for the client to select the version it actually
wants, other than whatever both support. IOW, it's not possible for the
client to say "I want version 8 but if that's not supported, use version
5".
All we do here is decide whether the connect event gets handled, clients
are always effectively connected (i.e. the client does send the connect
request) since we set up the backend during init.
Previously, a client could only bind to a capability immediately after
SEAT_ADDED and with a rather awkward API to confirm the capabilities.
Change this to allow for dynamic binding of capabilities, i.e. a client
calls ei_bind_capability() or ei_unbind_capability() whenever it feels
like, causing the respective devices of this capabilty to be added or
removed.
This allows for clients that are temporarily disinterested in a
capability but may require said capability later.
The default function takes one capability, a helper for enable/disable
multiple capabilities in one go is provided as well. On the protocol,
only the "bind" request exists which always represents the currently
wanted set of capabilities.
Note that the helper functions for multiple capabilities require NULL
(not zero) as sentinel, thanks to gcc.
This way eis_seat_has_capability() returns the effective capabilities
the server can actually use - no point creating touch devices when the
client has not confirmed that.
In theory we should have a eis_seat_get_effective_capabilities() to
differ between configured and effective capabilities, but I'm having a
hard time thinking of a use-case where the implementation forgets
which caps it enabled.
The side-effect of this patch is that adding a device without
capabilities requested by the client now produces warning.
Our API requires a client to know which capability to pass into the
drop_capabilities function. This doesn't work for capabilities newer
than the client's version so they do not get disabled. The client will
thus receive devices it didn't ask for and doesn't know how to handle.
Let's invert the requirement and require the caller to confirm the
capabilities it wants - all others are dropped.
This is an API break but also requires updates of all clients, the
previous simple case of just calling ei_seat_bind() will now result in
zero capabilities.
This effectively provides the EIS implementation with a notification
that the client will actually send events in the near future. To be used
by e.g. synergy-like clients when the pointer enters the logical screen
so that the EIS implementation can flash a warning or something.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
There's nothing in the protocol to modify the client device state from
the server, so a pause/resume cycle must leave the client with the
same(-ish) state. Pause is really just that, a short "no event now
please". Anything that would require e.g. modifying the device state by
releasing keys or buttons should result in the device being removed and
re-added.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This changes the protocol so that it is the EIS implementation that
creates devices within a seat.
A client now "binds" to a seat and the EIS implementation creates
devices matching the requested capabilities. A client can close a device
if it no longer wants those but otherwise everything (including pointer
ranges) is handled by the server.
This is one giant patch because changes at the protocol level cannot
easily be broken out into smaller patches. Some FIXMEs are left which
will be handled in follow-up patches, e.g. the keymap handling is
basically broken right now.
Fixes#7
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Removing a seat could cause two device remove events to happen. Fix this by
splitting the removal up into two bits: removed by server and removed by
client. Only once both bits are set, remove the device.
This needs to happen in libei and libeis.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
To cut down on the boilerplate, an unref-able struct variable can now be
declared as
_unref_(type) *name = NULL;
which is the equivalent of
_cleanup_(type_unrefp) struct type *name = NULL;
Let's see how that style ends up reading.
This means we can get rid of the custom _cleanup_foo_ functions everywhere, no
need for all the extra #defines etc. A somewhat special case is systemd which
defines the various unrefp functions for us in the headers, so we can use them
directly.
OBJECT_IMPLEMENT_UNREF now also creates the unrefp function for this object -
this of course conflicts where DECLARE_UNREF_CLEANUP_FUNC is in scope. Not a
problem so far, let's see how we go.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
After CONNECT, the EIS implementation needs to add one or more seats. The
libei client can only create devices within those seats. This mirrors the
wayland hierarchy as well as the X.Org one.
The seat has a set of allowed capabilities, so the client knows ahead of time
when it may not be possible to create a specific device.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Because events may be in-transit when a client removes the device, we need to
make this a full roundtrip to the server. Otherwise the client may assume a
device is removed, releases all references and then gets the original device
added event for that device. Better to have this as a round-trip instead.
This requires the server to call eis_device_disconnect() on the removed
notifications but we do so during eis_event_unref() anyway in case the server
forgets.
And it changes some of the API behaviors, so adjust the tests for that.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The previous approach would implicitly allow any capability not known to the
server. Switch instead to requiring an explicit 'ok' for any capability the
server wants to support.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Naming scheme is now: ei_device_<capability>_get/set_<what>. So far the
range is the only one where we had to deal with the same thing for two
different capabilities and it's unlikely we'll have to have different keymaps
for different capabilities. But still, let's do this now while it's still
easy.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Naming scheme is now: ei_device_<capability>_configure_<what>. So far the
range is the only one where we had to deal with the same thing for two
different capabilities and it's unlikely we'll have to have different keymaps
for different capabilities. But still, let's do this now while it's still
easy.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
In case the RESUMED isn't written on the wire yet (or it's written with
sendmsg and thus a separate message on its own) we don't have the RESUMED
ready for us yet. Let's call another dispatch to make sure it's processed.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Where the server connects a device after the client has already removed it, we
need to ignore that message silently.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This was already spelled out in the documentation but just not yet
implemented. New starting state for any device added by EIS is "suspended",
the server needs to explicitly resume it before events are accepted.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This function dispatches until a stable state is reached - that could be no
event waiting or some event waiting to be processed.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
