[RFC libevdev] Enable event logging for debugging

[Benjamin Tissoires]

On Thu, Apr 3, 2014 at 10:00 AM, David Herrmann <dh.herrmann at gmail.com> wrote:
> Hi
>
> On Thu, Apr 3, 2014 at 12:30 PM, Peter Hutterer
> <peter.hutterer at who-t.net> wrote:
>> This is a new approach to an earlier patch [1]. I'm running into more
>> use-cases where I really need the event sequence the caller sees to figure
>> out where the bug is. Note that the patch is just lightly tested, this is an
>> early RFC.
>>
>> I'll just copy the documentation here, that explains what I'm trying to do:
>>
>> + * libevdev checks the LIBEVDEV_LOG_EVENTS environment variable. If set to
>> + * "error", "info", or "debug", all events are logged with the respective
>> + * priority before they are passed to the client. This is useful to check
>> + * event sequences that happen before a specific bug in the higher layers of
>> + * the stack
>> + *
>> + * To debug specific devices only, the minor numbers of the device node may
>> + * be provided as a comma-separated list after the priority:
>> + * LIBEVDEV_LOG_EVENTS="error:1,2,3" logs all events from the devices with a
>> + * minor number 1, 2 or 3. The minor number comparison is performed on
>> + * libevdev_set_fd() and libevdev_change_fd().
>> + *
>> + * Since event logged by libevdev are likely to end up in some files and are
>> + * potentially attached to public bug reports, key event codes are
>> + * obfuscated to protect accidental leakage of private information. Key
>> + * codes between KEY_ESC up to including KEY_STOP are always logged as
>> + * "OBFUSCATED".
>> + *
>>
>> This should address some of the comments on the previous revision. It makes
>> it easier to debug a specific device only to cut down on the log spam. And
>> some of the security issues are alleviated by obfuscating key codes (I also
>> need to do this for MSC_SCAN) so that a password doesn't accidentally end up
>> in bugzilla somewhere.
>>
>> Comments?
>
> TBH, I don't really care. With structured-logging and journald, you
> can pass the device-node as ancillary data. This way, I can just use
> "journalctl DEVICE_NODE=eventXYZ" to filter for the device (or
> something similar..). But the patch looks fine to me.
>

Hmm, I don't like how these discussions usually turn into flamewar,
but here I must say that I believe libevdev is not tight to a systemd
environment. I really hope Android and Embedded Qt (those are the 2 I
can think of) will switch to libevdev and hopefully libinput. And I
doubt Android will rely on a systemd init.

So I am definitively in favour of being able to filter the device node
from libevdev directly.

Cheers,
Benjamin