Add BijectiveDictionaryBenchmark

[DandyLyons]

This PR adds benchmark testing using https://github.com/apple/swift-collections-benchmark/tree/main
Resolves issue #14

Context

Helpful resources:

1. https://github.com/apple/swift-collections-benchmark/blob/main/Documentation/01%20Getting%20Started.md
2. https://github.com/pointfreeco/swift-identified-collections/blob/main/Sources/swift-identified-collections-benchmark/main.swift

Usage

To run benchmarks you can use:

swift run -c release BijectiveDictionaryBenchmark run --cycles 3 results

To chart benchmarks you can use:

swift run -c release BijectiveDictionaryBenchmark render results chart.png   

For more info see: https://github.com/apple/swift-collections-benchmark/blob/main/Documentation/01%20Getting%20Started.md#running-benchmarks

Request For Comment

1. I think it would be helpful to compare at least some metrics to Dictionary. This can help us internally, as well as inform users of performance tradeoffs.
2. When ready, I think we should add a Performance section to the README, with a graph of the benchmarks. Similar to this: https://github.com/pointfreeco/swift-identified-collections/tree/main#performance

Room For Further Improvement

1. I will add benchmark tests for lookups, removals, etc.
2. In another PR, we should probably add CI.

Note

swiftLanguageVersions is deprecated in Package.swift after Swift 6. We should use swiftLanguageModes instead.

[DandyLyons]

Functions are now written directly inline in the benchmark tests to improve readability.

[DandyLyons]

Discovery: In the standard library, Dictionary has a suite of benchmark tests located at single-source. We can learn from these. They're using a different benchmarking library.

One thing I noticed is that they use @inline(never). It's possible that we should use this. However, I think it's already covered by the blackhole() method.

(You don't want to inline your benchmark tests because you want to guarantee that the compiler isn't artificially skewing your results by optimizing it.)

[DandyLyons]

There's still a lot of work to be done on the benchmarks. They haven't been thoroughly checked for correctness yet. But I though you would be interested in seeing some of the initial results.

[ladvoc]

I absolutely agree. The goal of benchmarking should be to understand and quantify the overhead BijectiveDictionary adds over Dictionary—both for making optimizations internally and for informing users of this library. Ultimately, I think it is a good idea to benchmark and compare all common operations between the two types.

I think we should first implement and test all key features, reaching feature parity with Dictionary, and then utilize benchmarks to make optimizations, using Dictionary as a performance baseline.

[ladvoc]

Discovery: In the standard library, Dictionary has a suite of benchmark tests located at single-source. We can learn from these. They're using a different benchmarking library.

One thing I noticed is that they use @inline(never). It's possible that we should use this. However, I think it's already covered by the blackhole() method.

(You don't want to inline your benchmark tests because you want to guarantee that the compiler isn't artificially skewing your results by optimizing it.)

Yes, it looks like this is covered by the blackhole() method and we safely omit @inline(never).