Make condition queries faster

[tonistiigi]

Initial benchmark shows ~700x improvement.

It doesn't use sandbox so its not 100% compatible with previous version but I think its very hard to accidentally leak anything. On the other hand it now allows combining conditions for some complex queries.

I started with a version that uses new Function() instead of eval. tonistiigi/node-bunyan@master...condition-function In my opinion its cleaner but in this version condition needs to be a Javascript expression. So instead on if() you would have to use ?: etc. (dtrace users should be used to this). Anyway, it also turned out to be slower than eval. The slow part is the assignment of __proto__ for the log level definitions, without it its faster than eval.

Benchmark results:

Tested on a real log file with 1k/30k rows. Conditions matched for ~5% of the rows.

Original 1k rows: 12.557s flamegraph
Eval 30k rows: 0.449s flamegraph
Function 30k rows: 0.551s flamegraph
Eval no dunder proto 30k rows: 0.424s flamegraph
Function no dunder proto 30k rows: 0.390s flamegraph

[trentm]

@tonistiigi Thanks for the spectacular analysis and I really apologize for the delay in responding.

I've implemented issue #87 in bunyan 1.0.0 using new Function. I need to look closer at your patch when I have the time again to consider eval. See also @nfitch's timing analysis of the equivalent -c ... open to my json tool at trentm/json#49