Previously we had UndefinedBehavior mixed with other sanitizers, in
special implicit conversions and overflows.
Also, run the sanitizers in Release configuration, so they will run
faster. -O1 is just needed for perfectly nice stack traces.
Also, moving the flags to an interface will prevent them to leak
to our external dependencies.
* [test-runner] Fix failed tests, generate android specific expectations
* [test-runner] Check render test runner testing result
* [test-runner] Add error case to android ignore list
* [test-runner] Update metrics
* [test-runner] Move Android render tests with big diff(>0.006) to ignore list
* [test-runner] Enable probe-testing android render test runner
* [test-runner] Add pixel4 baseline metrics
* Add git ignored metrics
* Make renderTestrunner run two times with different manifest config
- Try as much as possible not to use explicit compiler flags,
relying on CMake to do the job in a portable away.
- Build mbgl-core as a single static library and leverage lto as
much as possible.
- Support a MinSizRel build for other platforms.
- Make GLFW and Node a client of mbgl-core public interface.
- Same for the offline tool and static render tool.
- Use only target settings (do not pollute the global scope)
- Confine core code in a single file.
- Confine platform code inside a single include (exception being Qt
that needs AUTOMOC before adding the targets)
- Align the Android build with the Android toolchain file.
