Investigate automation — build-time or Edge networks?

[donmccurdy]

I've wondered for a while, when is the ideal time to apply optimizations in 3D assets deployed for the web? By necessity this is usually done by developers and artists now, often manually, and then the resulting optimized assets are checked into a repository.

There may be opportunities to make this easier, e.g. by hooking optimization tools like gltf-transform into web build tools like Webpack, Vite, Rollup, ...

Or perhaps build-time is not the right moment for some optimizations? These could be deferred to Edge networks (e.g. https://vercel.com/docs/concepts/functions/edge-functions) which would optimize the asset, then cache and serve it just in time. This approach allows some greater flexibility for frequent changes to the source asset, or device-specific optimization. For example, KTX2 textures could be pre-transcoded to before they reach the target device, based on information about the device GPU. This might involve support for Deno runtimes (#457).

There are just ideas, but I'd love to hear feedback, test some of these, and perhaps eventually "have an opinion" about the practicality of these automation choices.

[donmccurdy]

No current plans, other than #457.