Whats the difference between writing MethodChannel bridge versus dart:ffi bridge to run c/c++ code to get the response?

Question

Before I start my question I want to point out that it's not similar this question. Difference between writing platform specific code vs dart:ffi code. Here the questioner is asking the actual difference whereas I am trying to know the difference for the same task that can be achieved using both methods.

What's the difference in running native c/c++ code on the platform, getting the outcome on platform side(let's say Kotlin for instance) and sending it to dart via method channel versus writing the dart:ffi interface and directly calling the native c++ code. They basically will provide the same code execution. The only difference I see is that the MethodChannel call would be an async task vs dart:ffi which will be synchronous. Apart from the async behavior, will there be any difference(performance primarily) in getting the response from either of the technique.

Marcmarcano

Presumably it would be easier to code one ffi interface than all of Dart to swift and kotlin and thence to C through uni or whatever. Callbacks are a little easier with method channel over ffi. Presumably the ffi interface is more performant than serialising over a channel, but will it be significant? – Tribute

Answer 1

Here is a tour repo for dart:ffi: https://github.com/Sunbreak/native_interop.tour

1. Async/Sync

• For MethodChannel, both Dart -> Native and Native -> Dart are asynchronous
• For dart:ffi, Dart -> Native and Native -> Dart could be synchronous (except native call from non-mutator thread of Isolate)

2. Memory

• For MethodChannel, each interop requires serilization/deseriliazation
• For dart:ffi, you'd easily write C-like memory-efficient operation

3. Performance

dart:ffi synchronous interop has a good advantage on non-frequent small data

https://www.xdea.xyz/2020/11/flutter-platform-channel-%e6%80%a7%e8%83%bd%e6%b5%8b%e8%af%95/

Johnettajohnette

Does this mean that Method Channels are generally slower? What if I am trying to build an Engine that utilizes the device sensors, the GPU for rendering and inference, and also send requests to the cloud all in real time? Would a Method Channel be too much and ffi be faster? (with memory considerations) – Ballyhoo