About headless mode...

[mgerhold]

@kgnet88 provided some useful information on how it could be possible to run the engine in headless mode, but still provide OpenGL rendering (e.g. for unit tests). See their post here.

The main drawback I see here: Without providing (= paying for) a custom runner for our GitHub actions, we won't get any GPU support. Sure, there are projects that allow rendering OpenGL on a CPU, but I don't think this is worth the effort.

So I think: We should support headless mode (via a command line flag), but then the rendering system should be disabled alltogether. We can still support unit tests, but only for machines that have native GPU support and can open a window. I think that's okay.

[kgnet88]

If I understand you correctly, that means

• unit testing via hidden window (convenience) only on local machines
• headless server mode completely without graphics support
  correct?

I'd love to disagree with you, but I think you're right and it's the best solution for what you discussed in the stream at the moment.

The only scenario for a headless server mode with graphics support that I can think of at the moment would be to outsource calculations to a network node in order to increase the simulation / render power of the system. This is extremely interesting for me in the real-time simulation area, but I can't think of a good scenario for a game engine right now (except maybe cloud gaming, like Nvidia Now does).

I will definitely implement a few examples with the EGL and test them via a server, if only to try out how easy it is to outsource a physical simulation and how much overhead the whole thing produces, but of course that is far outside of your goals...

[mgerhold]

I don't even know if we need support for the "hidden window" approach because the software creates a window anyway (regardless of whether it's hidden or not). So for unit testing on local machines, I think "normal rendering" would be feasable as well. The software would only need support for saving screenshots.

As you pointed out: The only real use cases for headless rendering are way beyond the scope of this project, so let's try to keep it simple(-ish) 😅