use bufio for mse write

[trim21]

there is no need to start a new goroutine for writing small chunks.

If you want concurrently read write we can use chan, using slice, lock and cond to mock a chan doesn't sound like a good idea.

[anacrolix]

It's very old code, I'm sure I'd do it differently now.

Channels are usually over used in Go, they're also much worse for performance than people realise. I doubt this situation is affected.

Could you provide a benchmark to demonstrate the improvement?

Thanks for the contribution!

[trim21]

A local echo server.

before:
cpu: Intel(R) Pentium(R) Gold G6405 CPU @ 4.10GHz
BenchmarkMSE-4 4141 301669 ns/op 35309 B/op 220 allocs/op

after:
cpu: Intel(R) Pentium(R) Gold G6405 CPU @ 4.10GHz
BenchmarkMSE-4 4305 292442 ns/op 42028 B/op 207 allocs/op

[trim21]

emm I guess current implment is right. using bufio may cause a dead lock problem.

We should create a new goroutine to do the writing.

[anacrolix]

It's important that people test assumptions still hold true old code like this., so thanks for the effort and the conclusion.