On Jan 06 2017 or thereabouts, Peter Hutterer wrote:

On Wed, Jan 04, 2017 at 10:27:15AM +0100, Benjamin Tissoires wrote:

quoted

On Jan 04 2017 or thereabouts, Peter Hutterer wrote:

quoted

On Tue, Jan 03, 2017 at 02:33:36PM -0800, Ping Cheng wrote:

quoted

Hi Dmitry,

On Tue, Jan 3, 2017 at 1:30 AM, Benjamin Tissoires
[off-list ref] wrote:

quoted

On Dec 22 2016 or thereabouts, Dmitry Torokhov wrote:

quoted

On Mon, Dec 19, 2016 at 11:30:13AM +0100, Jiri Kosina wrote:

quoted

On Fri, 16 Dec 2016, Ping Cheng wrote:

quoted

Wacom Cintiq Pro [1] is a series of display tablets. They support
5 touch keys on the tablet. Those keys represent specific functions.
They turn display off; bring up OSD; bring up On Screen Keyboard;
bring up desktop control panel; and turn touch on/off.

The most left touch key, that turns display on/off, is controlled by
firmware. When the display is on, the mode is set. Otherwise, the
mode is off. So, it works like a switch. That's why we introduced a
new switch: SW_INDIRECT. The switch defauts to INDIRECT instead of DIRECT
was a request from useland, more specifically Gnome, developers.

Other four touch keys are true software keys. We use the existing
KEY_BUTTONCONFIG and KEY_CONTROLPANEL for OSD and control panel. But,
we have to request two new keys: KEY_ONSCREEN_KEYBOARD and KEY_MUTE_DEVICE
since none of the existing keys support those two actions.

[1] http://www.wacom.com/en-us/products/pen-displays/wacom-cintiq-pro

Signed-off-by: Ping Cheng <ping.cheng@wacom.com>
---
v3: Since no one has followed up with v2, let's add some more comments for
SW_INDIRECT so we keep the offlist decision public ;).

[ ... snip ... ]

quoted

diff --git a/include/uapi/linux/input-event-codes.h b/include/uapi/linux/input-event-codes.h
index d6d071f..32ef894 100644
--- a/include/uapi/linux/input-event-codes.h
+++ b/include/uapi/linux/input-event-codes.h

@@ -641,6 +641,9 @@
  * e.g. teletext or data broadcast application (MHEG, MHP, HbbTV, etc.)
  */
 #define KEY_DATA                 0x275
+/* same as SW_MUTE_DEVICE but triggered by a key */
+#define KEY_MUTE_DEVICE          0x278
+#define KEY_ONSCREEN_KEYBOARD            0x279

 #define BTN_TRIGGER_HAPPY                0x2c0
 #define BTN_TRIGGER_HAPPY1               0x2c0

@@ -781,7 +784,8 @@
 #define SW_LINEIN_INSERT 0x0d  /* set = inserted */
 #define SW_MUTE_DEVICE           0x0e  /* set = device disabled */
 #define SW_PEN_INSERTED          0x0f  /* set = pen inserted */
-#define SW_MAX                   0x0f
+#define SW_INDIRECT              0x10  /* set = not a direct input device */
+#define SW_MAX                   0x1f

I'd like to have explicit Ack from Dmitry on these ... Dmitry?

Sorry for the delay, but I think adding SW_INDIRECT is actually wrong.

What Wacom folks seem to need is way to re-classfiy the device (i.e.
update its properties) and let userspace know somehow. We can't keep
adding SW_INDIRECT and SW_NOTPOINTER and SW_NOTBUTTONPAD and so forth.
We however have EV_SYN/SYN_CONFIG that we may use to let userspace know
that device configuration changed and that userspace needs to
interrogate the device again. We can emit this code both when we change
properties as well as when we change ABS limits and changing keymaps and
so forth.

By "to interrogate the device again",  do you mean once we report
EV_SYN/SYN_CONFIG, userspace needs to reinitialize the device?

If not, how do we tell userspace which feature/info has been changed?

quoted

quoted

Benjamin, Peter, any opinion on the above?

Oooh, so that's the purpose of this event :) (the documentation says
"TBD"). I am fine with that. We would need to adapt the documentation in
Documentation/input/event-codes.txt and make sure the handlers are
properly adapted too (*maybe* add a SYN_DROP event to empty the queue of
the events in userspace).

Can you update the "TBD" in Documentation/input/event-codes.txt so we
have an agreed description to follow?

Here's a first attempt:

 * SYN_CONFIG:
   - Used to indicate that the device's static description has changed. This

Should we indicate the obvious fact that this is only after the device has
been registered and presented to userspace? I can imagine drivers adding
those notifications after each declaration while preparing the input
node :)

added, see below

quoted

quoted

     can happen when
     - at least one event type or code has been added or removed, or
     - at least one device property has been added or removed, or
     - at least one absolute axis has changed properties, or
     - the keycode has changed, or
     - the name, id, phys or unique identifier of the device has changed,
     A client should re-query the device once a SYN_CONFIG has been received
     as if the device was newly added. SYN_CONFIG does not indicate which
     information changed, a client is required to re-read the full 
     state. A SYN_CONFIG may be triggered even by types/codes/... that the
     client is not aware of, i.e. the state before and after may look
     identical to any given client.
     Dynamic values such as key state, switch state, absolute axis value,
     force feedback, etc. do not trigger a SYN_CONFIG. A SYN_CONFIG may
     trigger a SYN_DROPPED to terminate ongoing event sequences.

I think we should enforce this last sentence a little bit more:
"A SYN_CONFIG is basically a reset of the device so it should be
considered as a SYN_DROPPED from the client perspective. For backward
compatibility with clients not supporting SYN_CONFIG, a SYN_DROPPED
event is sent right before SYN_CONFIG."

Are we planning to force a SYN_DROPPED every time? Even when the event queue
is empty? I think it wouldn't do much other than keep the client busy. 

Come to think of it, there's a chance that this *reduces* backwards
compatibility. Until a SYN_DROPPED occurs, a client has no reason to ever
check bits again. But re-fetching all states during SYN_DROPPED is normal,
so if one axis disappears, the client may now get confused.

Ouch, very much ouch.

Maybe SYN_CONFIG should be limited to only props, axis ranges and new
axis?

quoted

Also, maybe we should add a warning about the udev properties:
"If a SYN_CONFIG allows to change the static description of a device, it
should be used with care. Some userspace pieces (udev) rely on the
device creation to tag it properly. Completely changing the device type
from a mouse to a touchpad is better handled through the destruction and
then creation of a new input node than relying on SYN_CONFIG"

(please fix typos/grammar/jibberish)

quoted

We should also add a blurb to the EVIOC*MASK ioctls about SYN_CONFIG.

Saying that it can't be masked and that it will invalidate the current
masks? Or this is the responsibility of the client?

I would just leave the masks as-is, less work and more predictability. This
means that a client has to re-apply them on SYN_CONFIG to avoid events from
new axes (if any) but that seems like the most straightforward approach. We
already allow masking out axes a device doesn't have, so invalidating
doesn't really give us any benefits. 

Ok, let's have a v2 draft:

* SYN_CONFIG:
  - Used to indicate that the device's static description has changed, after
    the device has been registered and presented to userspace.
    This can happen when
    - at least one event type or code has been added or removed, or
    - at least one device property has been added or removed, or
    - at least one absolute axis has changed properties, or
    - the keycode has changed, or
    - the name, id, phys or unique identifier of the device has changed,
    Do not send this event during the initial device initialization process,
    this event is exclusively for changes at runtime.

    A client should re-query the device once a SYN_CONFIG has been received
    as if the device was newly added. SYN_CONFIG does not indicate which
    information changed, a client is required to re-read the full 
    state. A SYN_CONFIG may be triggered even by types/codes/... that the
    client is not aware of, i.e. the state before and after may look
    identical to any given client.

    Dynamic values such as key state, switch state, absolute axis value,
    force feedback, etc. do not trigger a SYN_CONFIG. A SYN_CONFIG may
    trigger a SYN_DROPPED to terminate ongoing event sequences.

    SYN_CONFIG should be used with extreme care. There is no guarantee that
    a SYN_CONFIG event is handled correctly by a client, or even noticed.
    For example, udev relies on the original device description only to
    assign a device type. If a SYN_CONFIG were to change the device
    sufficiently to warrant a new device type tag, this may never be
    detected. Changing a device type (e.g. from mouse to touchpad) is better
    handled through the destruction and recreation of the input node.

I like it. Besides the legacy breakage we are talking about above.

Cheers,
Benjamin

Cheers,
   Peter