This way we have only one entry point, which ensures the device is ready
from the kernel point of view before we start injecting events.

Also add a few typing information and fixes here and there.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

We should not pre-instantiate the full class as it sets up udev and
whatnot. Keeping just the class type is enough.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

The new code is now way more reliable, but we never know.

Instead of manually retrying tests and messing up with the junit outputs,
use --retries to automatically retry them in the same pytest run.

And bump ci-templates while at it.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

This gives us a significant improvements in the tests for not a lot
of wasted CPU.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

This way we can switch backend. Instead of relying on udev, we can
use BPF to get the same information, but quicker.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Instead of parsing all devices one by one, we can rely on udev to only
know when the device has been initialized in the kernel through the "bind"
property.
Now we know that no more input nodes will be added.

Note: this rely on the fact that the module driving the uhid device has
been pre-loaded before the tests or that it was builtin the vmlinux

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

So we can have a `_BaseDevice` class completely without udev

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

We can check on the sysfs to get quite some information without having
wait for all the udev events.
This speeds up the creation of uhid devices without loss of functionality.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Not a lot of changes, but that helps.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

The objective is to strip out the udev requirement and speed things up.


Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Turns out that we wait 10 ms everywhere but here, and this reduces the
global time of tests by a fair chunk:
For test_wacom_generic.py, with 100:
```
==== 46 passed, 5 xfailed in 20.98s =====
```

With 10:
```
==== 46 passed, 5 xfailed in 16.38s =====
```

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Mostly asserts. There are still a few mypy errors, but we keep that for
later.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

self.current_report was never set, meaning that all of this code
was dead.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

The special marker case was ignored.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

turns out that there were no type checking for the input data, which
might blow up to the caller.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Those are the "easy" ones. No code is changing except for a few assert
to teach mypy that the value can not be None.

_HidRDescItem.dump_rdesc_lsusb() is removed as it was not used (and
wrong because "item = self.item() -> Object of type "str" is not callable"

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

I couldn't get the Match type to return either an str or a U16, so I
just ignored the type checking on lines 384 - 404.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

And ensure the CI is happy too

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

`black` is nice but for report descriptors it doesn't make sense to
reformat them, as the comments have a given signification

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

We are at kernel 6.1 now, almost 6.2, so we better test this one now.
As for flake8, it's always painful to run when black is so much simpler.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

The kernel expects an SPDX license instead of a full text blob.
Let's migrate the tests that are going to be synced with SPDX

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

The goal is to be able to easily sync that directory with the sources
of the kernel tree where the tests are going to live.

So by splitting the hid-tools specific tests from the device (kernel) ones
we make that way more easier to sync.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Compared to libinput, we are using master, not main. Keep main around in
case we switch to it.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Copied from libinput's CI

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Benjamin reported that the patches [1] added for wacom.ko testcases
had an issue. The input device exported to userspace is not caught
by the local udev rule, meaning that events are actually forwarded
to the session. This patch fixes that issue, as suggested by Benjamin.

[1] !128 (comment 1505073)



Signed-off-by: Ping Cheng <ping.cheng@wacom.com>

Signed-off-by: Roderick Colenbrander <roderick.colenbrander@sony.com>

This patch adds support for an additional DualShock4 feature report,
which is commonly used to retrieve the MAC address in USB mode.

Signed-off-by: Roderick Colenbrander <roderick.colenbrander@sony.com>

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Fixes couple of missing dependencies when installing via setuptools.

Signed-off-by: Tero Kristo <tero.kristo@linux.intel.com>

boot2container gives us several benefits over plain qemu:
- we don't start a full system which we need to ssh in
- we don't have to build a specific container
- we don't have to scp the files in and out

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

When running in an environment without udev, we might want to start
it manually.

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Jason Gerecke authored Sep 22, 2020 and Benjamin Tissoires committed Aug 11, 2022

The PTH-460 has its HID descriptor implemented slightly differently
to its siblings the 660 and 860. Specifically, each of the functions
of the pad are encapsulated in their own physical collections. This
causes a little bit of trouble for the current Wacom driver which
typically treats multiple collections of the same tool as multiple
snapshots of a single tool's state (mainly to support BT devices
that send e.g. 5 pen updates in a single packet).

We check to make sure that we don't get partial/empty pad syncs
when they shouldn't exist. The easiest way to trigger this behavior
is to simultaneously use the touchring and expresskeys since they
are in different collections.

This behavior was fixed in the upstream Linux kernel by commit
d4b8efeb46d9 ("HID: wacom: generic: Correct pad syncing").

Signed-off-by: Jason Gerecke <jason.gerecke@wacom.com>

Jason Gerecke authored Sep 21, 2020 and Benjamin Tissoires committed Aug 11, 2022

At the moment the function offers no middle-ground between
returning "true" and associating an evdev device with a specific
type, or returning "false" and associating it with *no* types.

The Wacom tests require us to have access to both e.g. Pen and
Pad devices, but we need to be careful to ensure that the pen
evdev device is actually associated with the "Pen" type, and
pad device with the "Pad" type. The "udev_input_event" function
does a pretty good job with this normally, but because all pad
devices are also pen devices, things go a little off the rails.

This commit allows the caller to return 'false' for only the
specific combinations of device and application that don't
make sense, but still return 'true' in other situations.

Signed-off-by: Jason Gerecke <jason.gerecke@wacom.com>

Jason Gerecke authored Sep 22, 2020 and Benjamin Tissoires committed Aug 11, 2022

Commit d4b8efeb46d9 introduced an issue which causes a BTN_TOUCH=0
event to be sent out whenever a heartbeat report is received. We
have a fix for this bug, but lets make sure it does not come back.

Signed-off-by: Jason Gerecke <jason.gerecke@wacom.com>

Jason Gerecke authored Sep 22, 2020 and Benjamin Tissoires committed Aug 11, 2022

Adds descriptors for the PTH-660, PTH-860, and PTH-460, as well as
the necessary device / test definitions to exercise them. Updates
`test_descriptor_physicals()` to better handle these tablets (new
usages listed among the required, filtering of mouse collection).
Also notes errata with the descriptors that cause the physicals
test to fail.

Signed-off-by: Jason Gerecke <jason.gerecke@wacom.com>