flatjson

flatjson

Fast and dirty parsing of JSON! Decode only the parts you care about!

What is flat JSON?

This library used to support only what I called a "flat" subset of JSON. But now it supports all JSON, but you can still decide how "flat" you want to go. The flatter, the faster :).

Flat JSON is a subset of JSON where the only supported types are objects containing strings, numbers, booleans or null values. There can't be nested objects or arrays. The root element must be an object.

What's the use for that?

It's fast and you decode only what you need. The old goal was:

If you log in JSON, likely your logs respect this principle. Using a JSON parser that supports only this subset should be faster than using a general purpose one. So this is one use case, parsing logs that are in JSON.

Implementation

This is an implementation of a flatjson parser. As far as I can tell, it works well enough. Check out the tests to see if its good enough for you.

Speed

Comparing this implementation with the standard library JSON decoder. Both have their output ignored:

• flajson's name/value pairs are ignored.
• encoding/json is decoding into an empty struct.

The goal here is to see how fast only the decoding part is. This is not necessarly a characteristic workload since a normal use case would allocate memory for the strings of the name/value pairs.

BenchmarkFlatJSON         1000000         1970 ns/op     177.15 MB/s           0 B/op          0 allocs/op
BenchmarkEncodingJSON     100000         20962 ns/op      36.83 MB/s        2151 B/op        130 allocs/op

At this time, the API of flatjson is not nailed down, so I haven't benchmarked a real use case. However, the benchmark above at least demonstrates that the potential for greatly improved speed is there.