The default is to include all artifacts, but we don't need any for
these jobs.

There appear to be some issues around those checks in CI jobs, and
some additional noisy output doesn't hurt there, so let's see what's
going on when it fails.

Before this change, the damage tracking debug flag enabled a mode
that displayed only the updated contents of the screen and filled
the rest with solid color. This may be useful when developing
wlroots' damage tracking, but it's not very effective for practically
anything else, since actually using the device in such mode is
close to impossible.

This commit changes it in a way that makes it mark the updated
screen areas with semi-transparent color rectangles. Yellow color
represents currently updated area, while magenta shows areas that
have been updated in a last frame (so single frame changes can still
be visually inspected).

Weston, sway and Mutter do it, and the protocol doesn't specify that
it shouldn't be sent.

Fixes gtk#7

Previously this was done only on successfuly frame copy, leading to
crashes.

It's 'long' gone. Completes 39c0729a.

This could happen by rounding to zero after scaling.

This makes it match the prefeferred coding style

Doesn't apply to XWayland views, as it's not easily achievable there
with X11 global coordinate system.

Phoc's startup is currently very quiet. Print the socket name on
startup. This helps to identify the socket when debugging but
also allows tests to fetch it.

Screencopy handlers were closely tied with single global PhocTestOutput.
Changing this will allow for easier handling of multiple outputs and
other sources of screencopy_frames, such as window thumbnails.

Implement window thumbnails via phosh-private+wlr-foreign-toplevel-management+wlr-screencopy protocol hybrid

Forward to 59f7adfe2977c3cb27a8def72c0d9fc2f2926c6d to fix compilation
on GCC 10.

See also: wlroots!68

Fixes #138

Since we're going to use it on the server side, it needs to stay in sync
with wlroots copy.

It allows us to use the GL calls directly for rendering.
GLES is already implicitly depended on via wlroots dependency.

This allows to reach back to the view that the handle represents.

It's going to be used by window thumbnails.

It's going to be needed by phosh-private.

We can turn this into a type check later on and it makes
sure log domains work correctly.

We handle virtual pointer there as well.

public. While at that rename to phoc_get_device_type ()

Instead of requiring a phosh/phoc lockstep upgrade post an error if a
client wants to bind the protcol. This keeps the code we need carry
minimal but we can still be backward compatible (this is not breaking
clients that don't use that part of the protocol).