Add the missing feedback flags to the Presentation extension protocol
specification.
These flags are slightly different from the previous RFCv3.1 definition:
http://lists.freedesktop.org/archives/wayland-devel/2014-March/013598.html
Now, all compositors are safe to use 0 as the flags if they don't bother
setting them properly. 0 is the "worst case" with the least guarantees.
The meaning of ZERO_COPY is not exactly the opposite of the old COPY
flag. ZERO_COPY is more strict, but applies only to that one surface.
Therefore it can be used to verify a zero-copy video playback pipeline,
also to a hardware overlay.
There is no longer a flag to clearly indicate if the final presentation
was done by a copy or a page flip. ZERO_COPY forbids the copy, but VSYNC
alone does allow copy in case it cannot tear. It is possible to have
first a compositing pass, and then another copy into the frontbuffer,
and still set VSYNC if it cannot tear. Usually "cannot tear" is too
hard to guarantee with a copy, so it often implies a page flip.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
Reviewed-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Tested-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Add accurate presentation timing features to Wayland: queueing and
feedback.
This specification is based on the draft written by Frederic Plourde
<frederic.plourde@collabora.co.uk> and redesigned by Pekka Paalanen.
The RFC v2 version is from
http://lists.freedesktop.org/archives/wayland-devel/2014-January/012988.html
Changes in v3:
* associate presentation time to current surface contents
This implements the suggestion from
http://lists.freedesktop.org/archives/wayland-devel/2014-February/013066.html
which prevents surface content from jumping backwards in time if a
client retroactively queues an update with a target time in the past.
* use 64-bit tv_sec in presentation
The time_t type used in struct timespec could be almost anything. POSIX
probably defines it to be an integer, but not the size. Apparently it is
usually 'long', which makes it 64-bit on x86_64.
To be able to fully represent timespec values returned by clock_gettime,
change the protocol to use 64 bits for the tv_sec part.
* define an error for invalid tv_nsec
This allow us to rely on the normalized timestamp form.
* define some interactions with sub-surfaces
Sub-surface cached state updates (synchronized mode) are designed
especially for resizing. As queued updates are not meant to produce any
resizing-like effects, they also do not trigger any sub-surface
operations.
* add sub-headings as xml comments
* queued update cannot map
Because before mapping, the surface has no main output assigned. An
immediate commit is needed anyway, to be able to set all the surface
state, which a queued update cannot touch.
* frame callbacks are not queued
It is not known when queueing frame callbacks would be useful.
Changes in v4:
* remove mentions of the queuing feature
The specification has been split and the queuing feature will be added
back in another version of the extension.
* add flags argument to 'presented' event
Describe the nature of how the update was presented to screen and the
characteristics of the feedback information. No flags have been
defined for now.
* add a protocol error code for invalid flags
Changes in v5:
* remove the destroy method for the feedback object
The protocol object should instead be automatically destroyed after
a 'presented' or 'discarded' event has been triggered.
* some grammatical corrections to the specification
[Louis-Francis Ratté-Boulianne: split the spec in two parts]
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
Signed-off-by: Louis-Francis Ratté-Boulianne <lfrb@collabora.com>
v3 Reviewed-by: Mario Kleiner <mario.kleiner.de@gmail.com>
