GitLab

May be nice to have for symmetry with the max value, but really, nothing
reads this value.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

We'll get that if we ctrl+c or unplug a device, let's just print it as
normal message, no need for a stack trace here.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

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

Together with better visual alignment
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

As the comment suggests, if a report (i.e. the event data) is larger than it
should be according to the ReportDescriptor, that's fine, we'll just end up
ignoring the bits.

This fixes issues with the stylus recordings seen in
libinput/libinput!558 (comment 792834)
That device sends 32 bytes for report ID 8 (13 bytes in the descriptor)

Fixes #32Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

I spent too long wondering why we fail a job but don't actually have any
explanation what failed or artifacts in this stage. So let's simply add a
message that points us to where those can be found.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

No point running through the array twice if we can just do it all in one go.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Where the file wasn't a human-readable descriptor, we'd end up returning an
empty list - and that is not None. So we ended up never detecting the format.

Regression introduced in 6de38774Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

These tests must be invoked via pytest, so let's remove the exec bit. It
wasn't set on all tests anyway.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Feature Reports can be const if they represent physical device properties,
e.g. Keyboard Physical Layout. Format those accordingly unless the Usage is
zero.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

In cases where we only have the HidReport object itself to pass around, it's
still useful to know which type that report is.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

bytes(x) gives us x bytes with value zero, making it a bit simpler than
decoding and ljust-ing a '\0' string.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

For input/output reports we don't really care about const items, they're
likely just padding. But Feature reports may have const items with
meaningful Usages, specified through min/max. For example my keyboard here
has Keyboard Physical Layout, Keyboard Form Factor, etc.

0x05, 0x0c,                    //  Usage Page (Consumer Devices)      56
0x0a, 0xc0, 0x02,              //  Usage (Extended Keyboard Attributes Collection) 58
0xa1, 0x02,                    //  Collection (Logical)               61
0x1a, 0xc1, 0x02,              //   Usage Minimum (705)               63
0x2a, 0xc6, 0x02,              //   Usage Maximum (710)               66
0x95, 0x06,                    //   Report Count (6)                  69
0xb1, 0x03,                    //   Feature (Cnst,Var,Abs)            71
0xc0,                          //  End Collection                     73

When parsing those, extract them as separate items, just like we do for the
variable data.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Slightly less confusing, I hope.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

On Fedora 33, the power_supply is not presented by udev with an 'add'
event, but a 'change'.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

We have a test that checks for the battery, but we present the device
whenever the inputs and leds are ready. This does not impact normally
because the battery is presented by the kernel before the power_supply
is added, but if we fail at adding the power_supply, we are failing
with an incorrect message.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

Printing the human form from an rdesc and re-reading that form should yield
the same report descriptor as before. For the test we use the same rdesc as
for the parsing test above, with some Logical Maximums fixed manually to
only use the number of bytes they should use.

Since we don't (yet) parse the byte number at the end of the human output,
we cannot guess that 0xff wants 2 bytes in the item.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

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

This is the format that hid-record spits out as part of the comments, e.g.

	Usage Page (Power Device)           0
	Usage (Vendor Usage 0x04)           2
	Collection (Application)            4
	 Usage (Vendor Usage 0x24)          6
	 Collection (Physical)              8
	  Usage (Vendor Usage 0x02)         10

We just search for at least one line with 'Usage Page (', if it exists then
we pass it to the parser.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

When parsing a human description, match the twos comp calculation from where
we generated the value to begin with.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

We shouldn't care about preceding spaces
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

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

This is the inverse of HidUnit.__str__(), generating the object from a
generated output string.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Allows us to go from a parsed HidUnit back to the 4-byte value it
represents.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Aside from better 'type safety' it moves all the code related to collections
handling/parsing into one location.

The new parsing code is more flexible than the old one too, previously we
only checked for a fixed prefix, the regex now is more flexible in regards
to whitespaces and how the hex code is written.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

It's unclear why this was handled this way but it produced wrong
calculations for items at the bit boundary - all those got bumped up to the
next size (e.g. vendor usage 0xffff ended up being 0x0000ffff in the report
because it got bumped to the next size).

The number of bytes needed is simply whatever the actual value requires, no
need to count the length of the hex string.

This produces the wrong value for cases where extra bytes are present in the
original file descriptor, e.g. this case:
  # 0x26, 0xff, 0x00,              //   Logical Maximum (255)
This line reconverted would output a single-byte only
  # 0x25, 0xff                     //   Logical Maximum (255)

Without parsing the byte count at the end of the outputted line, this isn't
something that can be fixed here.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Unit(None) is a value of 0, so let's get on with it. This should be handled
correctly by the normal parser, but until that one is fixed just make this
one a special case.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

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

All predefined systems have None at index 0
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Before: Collection (Vendor_defined)
After: Collection (Vendor_defined 0x81)

This matches our behaviour for vendor usages/usage pages and is required to
recover the exact bytes from the human description.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

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

The comments we print will append the byte in the line, so we end up with
this:

   End Collection           123

End Collection doesn't have parenthesis, so it's not covered by the existing
condition. Let's handle it separately.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

For those strings, the value is the 0xd0.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

parse() is too flimsy when it comes to handling optional whitespaces, let's
replace it with regex matching.

This now correctly handles the optional exponent and the case of Unit(None)
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

In case of a failure, it might be interesting to have a peek at the
console log. Export it as an artifact.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>

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

Allowed to fail because we're not tracking the upstream kernel so some tests
are expected to fail. But this will at least detect severe breakages that
pass the existing flake/setup.py checks.

Before we can run the test though we need to modprobe all of the hid modules.
The ITE keyboard has its own kernel module and that causes device creation
to fail, specifically: "It turns out that this device loads the hid_ite
kernel module. So what happens is that we create the hid device, hid-generic
sees it, presents the nodes to the userspace, but then udev loads hid-ite,
which kills the input nodes attached to hid-generic and everything related
to it. The next attempt at loading the device works fine, because hid-ite is
there already."
See #29 (comment 766562)
Simply modprobing all modules before we run the test fixes this issue. This
is a separate script to avoid the escaping of subshells otherwise needed to
get to the kernel module path.

Fixes #29Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Prep work for enabling qemu images - ci-templates does not support those for
arch yet so Fedora it is.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

Create an (arch) image and use that to run our jobs. For now this primarily
avoids falling afoul of the docker pull limits. No real functional changes,
at least while the Arch image is recent.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>

We don't need udev rules until we run a test that creates a test device.
So we modify our class to set up the udev rule on the second context
entered (first is session, second is the current test case) and clean up
once the last context (== session context) is exited.

This way we never mess with udev rules if we only run tests without test
devices.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>