mscdex

> Not if you are piping > (unless this is implementing in the piping too) I don't follow. If you had `socket.push(data)` inside the `onread` callback and had the `buffer`...

> The way I understand this part of the code, is that this buffer is static and it is reused for every write? Not necessarily, the `buffer` property can also...

> This would leave it to the user to scale the buffer? Or do you propose that the same mechanism is implemented in the JS-part of the library? I would...

The [`net` benchmarks](https://ci.nodejs.org/view/Node.js%20benchmark/job/benchmark-node-micro-benchmarks/1042/console) look fine to me. I wonder if this will have a noticeable impact on memory usage one way or the other in practice?

> * Should we break bad applications that are used to receiving 64k per read Seeing as how we're dealing with streams here, nobody should be relying on chunk sizes.

Getting a lot of unrelated EACCES failures for parallel/test-fs-rm on multiple platforms.... not sure what that's about. EDIT: I'm guessing f9447b71a6b458adbb265f8a5fd38ebf41bc97a4 may be the culprit for that

Where is SM2 used?

I would vote to just leave it out for now and if there is a pressing need to add it in the future we can revisit it then, especially considering...

Of the choices, I'd vote for option 3.

These changes appear to have some fairly large performance regressions: ``` assert/ok.js n=100000 *** -77.20 % ±2.93% ±3.94% ±5.21% assert/throws.js method='doesNotThrow' n=10000 *** -41.40 % ±5.50% ±7.37% ±9.71% ```