Mixing tokio-util's Framed with raw Bytes buffers

[locryus]

I have a protocol which is implemented very nicely with Framed, apart from one annoying detail: Many of the frames carry a payload, which can be pretty large (up to 256KB currently, but may be bumped to 1MB).

These payloads are Bytes buffers which are already loaded into memory at the point the frame is created. I've been copying the payload into the Encoder's BytesMut, which works, but the application is sometimes copying multiple TB of payload every day, and it hurts my soul to perform all those "unnecessary" copies.

For such protocols, is there a way to retain the convenience of Framed, but also not have to perform all the copies?

If not, is there any non-breaking way that this could be supported?

Ignorantly spit-balling here: I suspect that the the Encoder implementor would need a new method that supports accepting a Bytes buffer, and the Framed internals would need to keep a list of potentially mixed BytesMut and Bytes elements.

As a work-around, if any of the above is unfeasible: How costly, and foot-gunny, would it be to deconstruct the Framed into its inner parts (and later reassembled) each time a payload needs to be sent?

[Darksonn]

You should probably open code Framed in your codebase to make the modifications you need. I wouldn't try to use the existing abstraction if it becomes too hard to do so.