Best practices on exposing Bevy data to the outside world?

[barren yacht]

I am trying to expose data from my Bevy app to non-Bevy callers within the same address space without going through disk I/O.

The general idea is to use Bevy as the backend to run a simulation and periodically query it to extract state. Kinda similar to what the Render World is doing, but at random rather than each frame, and probably via a DLL.

The data will be Plain Old Stack Data owned by the receiver, i.e. we don't need to worry about lifetimes or anything; I'm pretty flexible on the structure(s). It doesn't need to be generic, the number of types exposed will be finite.

So far I've gotten comms going the other way by creating a static ⁨AtomicBool⁩ that the DLL user can toggle, so that seems like one option - but I was wondering if anyone might have a neater pattern around this?

[ionic elbow]

If you aren't using winit, you can set your own app runner and have it do something with the state and/or store it somewhere when your conditions are met
https://docs.rs/bevy/latest/bevy/prelude/struct.App.html#method.set_runner

You can also use a crossbeam channel (or a pair of channels for two way comms) https://docs.rs/crossbeam/latest/crossbeam/channel/index.html

[drowsy whale]

another option is for the DLL to provide a callback, and then having a bevy system call that callback with the data. the advantage would be no need to use a channel, the disadvantage would be the external code can't ask for updates, but has to wait for the next callback

[barren yacht]

That could actually be pretty nice in some ways, a push model rather than a pull model.

I'd need to dig up some examples of calling a function pointer via the FFI from Rust though, since I have no idea what that looks like rn.

I'm concerned it might be a bad deal in this case - I don't really need the flexibility of a callback here and might be more brittle - but I think it's worth keeping in my back pocket for the future

[drowsy whale]

yeah if your data is cheap to clone then a channel will work well

hell, if your data is really cheap and access infrequent you could just use Bevy Remote Protocol and not need linking at all

[barren yacht]

ooooh

I've seen BRP and entirely forgot it can do these kinds of things

🤦‍♂️

[drowsy whale]

hey if that works for your usecase that's really cool

[barren yacht]

requires a SerDe into JSON, which is a bit painful

[drowsy whale]

thats why i said infrequent access, but its not actually that slow

fast enough for an entity inspector

[barren yacht]

Probably too slow for my purposes, it's AI code

I guess if you only used it for the coarse stuff and batch it, might be fine