Improve performance of pretty-format

[jamiebuilds]

This doesn't support everything yet, but it is a pretty big restructuring that will allow it to work much much faster.

See comparison: https://gist.github.com/thejameskyle/fce69285d4682923061bf5f83461cc7b

cc @cpojer

[cpojer]

Nice work!

[jamiebuilds]

Hmm– testing with really large objects this is actually quite a bit slower. 9930672ns vs 6715812ns

[jamiebuilds]

Never mind, I realized I've been doing something stupid and measuring this completely wrong. Here is the corrected comparison: https://gist.github.com/thejameskyle/f8b50bf559360549326b52f3bb323d5d

[cpojer]

Not sure I can trust you again after this.

[jamiebuilds]

Not sure I can trust you again after this.

Hey I improved the performance without even having real metrics to guide me. That must count for something

[jamiebuilds]

Okay, got everything working again and here are the new perf results: https://gist.github.com/thejameskyle/2b04ffe4941aafa8f970de077843a8fd

[cpojer]

I love it. Fantastic work, @thejameskyle.

[jdalton]

Woot!

Heads up! There is some handy process.binding('util') helpers for isTypedArray, isMap, isSet, etc. that are ~2x faster than that of traditional Object.prototype.toString.call checks (Lodash will start using them). Also, engines have made Array.isArray pretty fast.

[jamiebuilds]

nice, thanks

[jdalton]

No prob. I've opened an issue on Node to propose exposing more of them ⚡