UTF-8 isn't encoded/decoded correctly about msgpack-lite

[kawanet]

0.1.26 published. Thanks!

[maciejhirsz]

Note: benchmarking things on my own, I found that building strings by simply concatenating to a string via += operator is faster than creating a preallocated array and calling String.fromCharCode.apply.

[kawanet]

building strings by simply concatenating to a string via += operator is faster than creating a preallocated array and calling String.fromCharCode.apply.

Really!?!? I was pretty surprised with this!

https://github.com/feross/buffer/blob/master/index.js#L997
Buffer shim is also use the same String.fromCharCode.apply technique still.

I guess that buffer-lite.js on the master branch creates two Arrays at least on toString. It may cost. Building/parsing string would give the most impact for decoding/encoding performance. I'd love to see it in depth.

Thanks for the great patch anyway.

[maciejhirsz]

Note: the benchmarks I was running for this were on Node 6.3 and Chrome Canary 54, so take it with a big grain of salt, different browsers might vary.

For longer stings (>~128) if your buffer happens to be Uint8Array, the absolutely fastest string decoding you can get is by passing a Uint8Array#subarray to TextDecoder#decode - it's comparable in performance to using Buffer#toString in Node.js, and beats JSON.parse. The threshold is whatever the cost there is to initialize a new TypedArray, which seems to always be in the range of 100 - 500 ns on my hardware for whatever reason, even when using already allocated ArrayBuffer.

[gordomium]

Hope it be fixed