I'm reminded of protothreads (http://dunkels.com/adam/pt/). I used protothreads some time ago on an embedded project to implement a (very small) HTTP server in only a few KB or RAM.
In my experience of using protothreads, only the functions using the protothread macros need to be written in a special way:
- passing a context object
- being careful of variable initialization/blocks
in order not to work on unitialized memory.
Why would functions called by "lightweight threads"/"coroutines"/whatever want to yield? One example is, they want to wait for an event - an I/O request completion or a computation offloaded to an accelerator or whatever - without having to return all the way back to the event loop.