Add derived vs. non-derived benchmark

[travisbrown]

I've been meaning to add a benchmark comparing derived and non-derived codecs for a while, and this question from @adelbertc was a good reminder.

The non-derived decoders and encoders are intended to be idiomatic, and one of the interesting things the benchmarks show is that Json.obj (used in the non-derived encoder but not the derived one) needs some work—the derived encoder is actually faster than the hand-written one. The throughput of the derived decoder is about 3% lower than the non-derived one, which is within the margin of error.

Benchmark                                   Mode  Cnt        Score       Error  Units
CirceDerivationBenchmark.decodeDerived     thrpt   80   788832.569 ± 11653.684  ops/s
CirceDerivationBenchmark.decodeNonDerived  thrpt   80   813443.980 ± 13189.326  ops/s
CirceDerivationBenchmark.encodeDerived     thrpt   80  1010433.746 ± 12266.961  ops/s
CirceDerivationBenchmark.encodeNonDerived  thrpt   80   975425.816 ±  9003.887  ops/s

The derived decoders and encoders also look pretty good in terms of allocations:

Benchmark                                                       Mode  Cnt        Score       Error   Units
CirceDerivationBenchmark.decodeDerived:gc.alloc.rate.norm      thrpt   20     4984.002 ±    35.636    B/op
CirceDerivationBenchmark.decodeNonDerived:gc.alloc.rate.norm   thrpt   20     4844.002 ±   110.473    B/op
CirceDerivationBenchmark.encodeDerived:gc.alloc.rate.norm      thrpt   20     2752.002 ±     0.003    B/op
CirceDerivationBenchmark.encodeNonDerived:gc.alloc.rate.norm   thrpt   20     2992.002 ±     0.003    B/op

So performance probably isn't usually a good reason to avoid generic derivation here.

[codecov-io]

Current coverage is 70.06%

Merging #96 into master will increase coverage by +0.79% as of 967a10e

Powered by Codecov. Updated on successful CI builds.