Investigate the use of cesium-native in a WebAssembly build

[kring]

One of the potential future uses of cesium-native - though largely unexplored so far - is to compile parts of it to WebAssembly and incorporate it into CesiumJS and potentially other in-browser rendering engines.

Meanwhile, we haven't been bashful about using C++17, STL, and third-party libraries where they are helpful, all of which could potentially bloat a WebAssembly build beyond the point where it is practical to use it in this way. On top of that, some features, such as C++ exceptions, may be problematic in WebAssembly builds.

There is a tradeoff here, between moving quickly now by using the best tools at hand and not worrying excessively about the future impact on a WebAssembly build, and possibly needing to make major changes to cesium-native in the future if and when we do need to build it for WebAssembly.

For now, I don't think we'll do anything drastic, but will:

• Avoid (but not outright forbid) the use of exceptions.
• Select third-party libraries carefully. Don't use Boost.
• Deal with most of the consequences later.
• Perhaps it'd be worth spending a little time doing a WebAssembly build just to see where we stand?

CC @jtorresfabra @pjcozzi

[pjcozzi]

🙏 @kring

CC https://github.com/CesiumGS/product/issues/28

[javagl]

One bullet point from the linked issue was

• Can we make the 3D Tiles loader generic, tiny3dtiles, like https://github.com/syoyo/tinygltf

This made me wonder whether there is a smaller, more basic library hidden in the current Cesium3DTiles. Maybe something that is (roughly) only an "in-memory representation of 3D Tiles", without the infrastructure for network requests, asynchronous operations and rendering...? (The latter is where much of the compexities of threading and exceptions comes in...)

[jtorresfabra]

I would say the sooner we start to test, the better. There are a lot of things to test/do, not only binary bloating, but also if the multithread approach works in wasm, memory consumption, compiler optimizations, no need for rtti/exceptions, image decoding intercept (as stb_image is awfully slow compared to browser image decoding), etc.

Let me know if I can be of any help, happy to try to make a build for wasm if I get some time to work on it. CC: @pjcozzi @lilleyse

[kring]

I changed the title to make it clearer how this is actionable.