This implements client side window decorations of our windows. The title
bar of our windows is now drawn by Android and we simply map out the
title bar area hit test area to SDL to detect when the other wants to
move the window or resize it.
dbus-cpp was dropped from 18.04 as it is not maintained anymore so we
should migrate to something more stable with a long term perspective.
libsystemd fits in nicely here as it has a nice and clean dbus
implementation which is battle tested and can be easily integrated into
our existing C++ code base.
This allows building against the same version of GLM across all targets
and not deal with API changes of newer GLM versions which introduce
regressions. Long term we want to migrate away from GLM and stick to a
minimal self-written subset.
Newer Ubuntu systems tend to not provide a libGLESv1_CM.so.1 anymore. As
we're not required to have a GLESv1 implementation (only a few older
Android apps are using GLESv1) we can as easy fix avoid loading a real
GLESv1 implementation and assign dummy functions instead to all relevant
GLESv1 API calls. In the future we want to switch to a GLESv1->GLESv2
translator as the Android emulator does.
OpenGL ES is enough for our purpose and we don't want to maintain a
unneeded OpenGL ES to GL translator. All target drivers have proper
OpenGL ES support these days so no need to depend on GL anymore.
If we expose too many extensions we don't support ourself yet (like
EGL_ANDROID_native_fence_sync) then we get unexpected corruption or
missing screen updates in our rendered output.
Fixes#635, #457, #692, #582 (and a few more duplicates)
`l` is a format for `long int`. It accidentally worked on 64 bit architectures, because `long int` has the same length as `size_t` there.
It didn't work on 32 bit architectures, where `size_t` is shorter than `long int`.
Let's use `z`, because that's the proper format for `size_t` -- it works on any platform.
If container manager dies before session manager dying,
the session manager cannot exit if it waits for all.
This patch fix the above issue.
Signed-off-by: Han Pengfei <hanpfei@gmail.com>
It is possible for more than one message to be delivered concurrently.
When this happens, messages are concatenated together (end-to-end).
Current code errors out if the total received data is larger than the
known header_size (4) + the body size (reported in the header).
However, when >=2 messages arrive at the same time, the total received
data will be far greater than this.
The current code base actually handles multiple messages perfectly
already, so all we need to do is remove the early return and
everything "just works".
Signed-off-by: Lee Jones <lee.jones@linaro.org>
This will make the *.entries files the the authoritative place to
define additional GL function prototypes. All generated header
files are now placed inside the build directory so that nobody
will ever try to touch them again.
