initial survey on headless render/testing for WebGL [uts-001N]

✍️source

I want to be able to run headless render or visual regression tests, both locally where GPU is available, and on CI where GPU has to be emulated.

The ingredients seems to be Storybook+Playwright+Xvfb.

- use OrbStack as local docker engine
- use act (usage) to run GitHub Actions locally
- use Storybook to building pages and components in isolation
- use Playwright to verify the visual appearance on different browsers
- use Xvfb to emulate a display server

If I'm a bit more ambitious, I would like to be able to do the same for WebGPU. At least we do know that Xvfb support OpenGL via mesa (which also supports Vulkan etc.), and there are some libraries to run WebGPU via WebGL 2 (e.g. wgpu).

Some resources I found so far:

- Effortless CI Integration: Running Playwright Tests with GitHub Actions uses `xvfb-run`
- `playwright`'s own CI uses `xvfb-run`
- Headless chrome – testing WebGL using playwright demonstrates WegGL testing when GPU is available, see also its writeup
- Options to enable GPU Hardware Acceleration in headeless mode
- use xquartz with PyVirtualDisplay on Mac
- use libosmesa with Chrome from 2016
- use VirtualGL with Xvfb ⧉