[PATCH] implement weston local texture using weston_buffer without wl_resourse

[Pekka Paalanen]

On Sun,  8 Mar 2015 10:33:25 +0100
Benoit Gschwind <gschwind at gnu-log.net> wrote:

> This patch allow desktop-shell and other components of weston to create
> local texture as surface and show them. This patch support only the
> gl-renderer.
> ---
>  Makefile.am                |  2 ++
>  src/compositor-drm.c       |  3 ++
>  src/compositor.c           | 51 +++++++++++++++++++++++++++---
>  src/compositor.h           |  9 ++++++
>  src/gl-renderer.c          | 77 +++++++++++++++++++++++++++++++++++++++++++---
>  src/weston-local-texture.c | 75 ++++++++++++++++++++++++++++++++++++++++++++
>  src/weston-local-texture.h | 56 +++++++++++++++++++++++++++++++++
>  7 files changed, 264 insertions(+), 9 deletions(-)
>  create mode 100644 src/weston-local-texture.c
>  create mode 100644 src/weston-local-texture.h

Hi,

recapping and extending on what I said in IRC about this general idea:

As far as I understand, local buffers are only usable for drawing
happening in the compositor process. Drawing in the compositor process
is a short-cut that avoids several synchronization and IPC complications
compared to a helper client.

Traditionally we have made the policy decision to not do any drawing in
the compositor process apart from compositing. Drawing takes time, and
during that time the compositor is blocked. Drawing also likely means
loading more and complicated libraries to the compositor process, and
increasing the risk of bugs. (cf. If weston-desktop-shell crashes,
it'll respawn so fast you probably won't even notice.)

However, we already also have an exception to this in upstream: the
Wayland backend needs to draw the output window decorations.

Personally I'm not fond of the idea of adding explicit support for
local buffers, but I understand it may be very attractive to smaller
DEs that are based on Weston. Maybe when we have libweston, this
feature addition would see good use. Maybe one could even convert the
Wayland backend to use local buffers.

Therefore, I will not review this feature at this time. If there are
other people who want to review this and work on it to get it merged,
I'm ok with that and won't object as long as it meets our quality
standards.


Thanks,
pq