The overhead is very limited. The memory overhead is fixed, and the compute overhead is linear with the rate at which signals arrive (which is typically super

The overhead is very limited. The memory overhead is fixed, and the compute overhead is linear with the rate at which signals arrive (which is typically super low). Plus the whole thing is very easy to reason about. And you never have to worry about masking signals that you handle with a self-pipe. Add in the portability, and it's a very clear win.