Adopt ISO8601FormatStyle on newer platforms

[czechboy0]

Motivation

ISO8601FormatStyle is significantly faster than ISO8601DateFormatter.

Proposed solution

Use #available to dynamically use the faster and more modern implementation in ISO8601FormatStyle, fall back to ISO8601DateFormatter otherwise (today's behavior).

We should also add benchmarks to the runtime package while we're there to measure the improvement.

Alternatives considered

No response

Additional information

No response

[arthurcro]

Hey! I would be happy to help with this if it's not already picked up! 😊

[czechboy0]

Thank you @arthurcro - go ahead! 👍

[simonjbeaumont]

@arthurcro thanks for getting stuck in! Just a note that we don't currently have benchmarks in the repo, and we're about to migrate the CI to GitHub Actions.

This shouldn't block you working on the issue, but it might delay us landing it, as we'd like to get this in place. I'll try and schedule that work this week.

[simonjbeaumont]

@arthurcro I expedited this to ensure you're unblocked to work on this when you get to it. The CI is updated. We have an open PR to add package benchmarks: apple/swift-openapi-runtime#119.

And it looks like this should be a good patch...

========================
OpenAPIRuntimeBenchmarks
========================

Date.ISO8601Format(_:)
╒═════════════════════════╤═════════╤═════════╤═════════╤═════════╤═════════╤═════════╤═════════╤═════════╕
│ Metric                  │      p0 │     p25 │     p50 │     p75 │     p90 │     p99 │    p100 │ Samples │
╞═════════════════════════╪═════════╪═════════╪═════════╪═════════╪═════════╪═════════╪═════════╪═════════╡
│ Malloc (total) *        │       2 │       2 │       2 │       2 │       2 │       2 │       2 │       5 │
├─────────────────────────┼─────────┼─────────┼─────────┼─────────┼─────────┼─────────┼─────────┼─────────┤
│ Time (total CPU) (ns) * │     771 │     773 │     773 │     782 │     794 │     794 │     794 │       5 │
╘═════════════════════════╧═════════╧═════════╧═════════╧═════════╧═════════╧═════════╧═════════╧═════════╛

ISO8601DateFormatter.string(from:)
╒═════════════════════════╤═════════╤═════════╤═════════╤═════════╤═════════╤═════════╤═════════╤═════════╕
│ Metric                  │      p0 │     p25 │     p50 │     p75 │     p90 │     p99 │    p100 │ Samples │
╞═════════════════════════╪═════════╪═════════╪═════════╪═════════╪═════════╪═════════╪═════════╪═════════╡
│ Malloc (total) *        │       9 │       9 │       9 │       9 │       9 │       9 │       9 │       5 │
├─────────────────────────┼─────────┼─────────┼─────────┼─────────┼─────────┼─────────┼─────────┼─────────┤
│ Time (total CPU) (ns) * │    1536 │    1547 │    1566 │    1574 │    1606 │    1606 │    1606 │       5 │
╘═════════════════════════╧═════════╧═════════╧═════════╧═════════╧═════════╧═════════╧═════════╧═════════╛

ISO8601DateTranscoder.encode(_:)
╒═════════════════════════╤═════════╤═════════╤═════════╤═════════╤═════════╤═════════╤═════════╤═════════╕
│ Metric                  │      p0 │     p25 │     p50 │     p75 │     p90 │     p99 │    p100 │ Samples │
╞═════════════════════════╪═════════╪═════════╪═════════╪═════════╪═════════╪═════════╪═════════╪═════════╡
│ Malloc (total) *        │       9 │       9 │       9 │       9 │       9 │       9 │       9 │       5 │
├─────────────────────────┼─────────┼─────────┼─────────┼─────────┼─────────┼─────────┼─────────┼─────────┤
│ Time (total CPU) (ns) * │    1528 │    1529 │    1530 │    1541 │    1558 │    1558 │    1558 │       5 │
╘═════════════════════════╧═════════╧═════════╧═════════╧═════════╧═════════╧═════════╧═════════╧═════════╛

As part of the work, it would be good to remove those benchmarks against foundation, once we have comparable numbers for the ISO8601DateTranscoder.

[arthurcro]

Thank you @simonjbeaumont, I appreciate it! I will start on this issue as soon as possible. I'll make sure to also update the benchmarks and remove the ones against foundation.

[arthurcro]

Hey @czechboy0, I made an early draft MR on my fork but while running the tests I noticed testDateTranscoder_iso8601WithFractionalSeconds is failing with:

testDateTranscoder_iso8601WithFractionalSeconds(): XCTAssertEqual failed: ("2023-01-18T10:04:11.122") is not equal to ("2023-01-18T10:04:11.123Z")

I tried reproducing the issue in a Swift playground, and got the same results. Also, parsing the formatted date with the same Date.ISO8601FormatStyle gives me a different date.
I wonder if that's expected or might be a Foundation issue. 🤔

let initialDate = Date(timeIntervalSince1970: 1_674_036_251.123) 
let formatStyle = Date.ISO8601FormatStyle.iso8601
    .year()
    .month()
    .day()
    .time(includingFractionalSeconds: true)

let encoded = initialDate.formatted(formatStyle)
print(encoded) // prints 2023-01-18T10:04:11.122 instead of 2023-01-18T10:04:11.123
print(initialDate == (try! formatStyle.parse(encoded)) // prints false

[czechboy0]

@arthurcro that's interesting, can you open an issue against https://github.com/swiftlang/swift-foundation if you manage to reproduce it on a minimal project that doesn't include Swift OpenAPI Generator?