SR-11766: Fix coefficients for UnitVolume

[ole]

This fixes Measurement values expressed in cubic centimeters and cubic millimeters.

I found the errors in the coefficients because I tested one of my libraries on Linux for the first time and a test was failing (see ole/Ampere#25).

I considered adding unit tests that test some conversions to Foundation, but I'm not sure how useful they would be. A test for converting a Measurement from one unit to another would essentially repeat the code that's being tested; it seems it would be almost equally likely for a bug to be in the test as it would be in the code being tested.

[ole]

Please let me know if you'd consider tests of this kind useful:

func testVolumeConversions() {
    let input = Measurement(value: 5, unit: UnitVolume.liters)
    let megaliters = input.converted(to: .megaliters)
    XCTAssertEqual(megaliters.value, 5e-6, "Conversion from liters to megaliters")
    let kiloliters = input.converted(to: .kiloliters)
    XCTAssertEqual(kiloliters.value, 0.005, "Conversion from liters to kiloliters")
    let deciliters = input.converted(to: .deciliters)
    XCTAssertEqual(deciliters.value, 50, "Conversion from liters to deciliters")
    ...

[spevans]

@swift-ci test

[spevans]

@ole It would definitely be good to get some tests added, currently there dont seem to be very many. If you could add a couple that would validate the constants used for conversion that would be great. Also, don't worry about the macOS test failure, thats an unrelated issue.

[spevans]

@swift-ci test macos

[ole]

@spevans Thanks. I will add tests, but I don't know when I'll get to it. Hopefully sometime next week.

[parkera]

Thank you @ole !

[ole]

@spevans @parkera I added some tests for UnitVolume conversions. Let me know what you think. I tried to use conversions that readers of the tests would find reasonably intuitive so that mistakes have a chance to be spotted (e.g. it's more likely that readers will know there should be 4 quarts in a gallon than that 1 gallon should be x.xx cubic inches).

[ole]

Admittedly, this assertion with the insanely high accuracy threshold is kind of ridiculous. I wanted the test to be intuitive, so for people who know that 1 mile = 1760 yards, it makes sense that 1 cubic mile = 1760^3 cubic yards.

Unfortunately, that's not really the case with Measurement (even after factoring in floating-point rounding errors) because the conversion factors hardcoded by Foundation often don't use the full range offered by Double. In this instance, cubicMiles is defined like this (in Unit.swift):

static let cubicMiles           = 4.168e+12 // in liters

Whereas the "accurate" conversion factor is more like 4.16818182544058e12 (= 16093.44^3; 1 mile is 16093.44 decimeters) (I think), which leads to the huge discrepancy (in absolute terms) between the actual and expected values.

This inaccuracy is not specific to corelibs-foundation, but is the same in Darwin-Foundation. Out of interest, was it a conscious decision to limit conversion factors to x significant digits or something when the Measurement types were introduced?

[spevans]

As far as scl-foundation is concerned it just tries to match Darwin's version as closely as possible, I don't know why Foundation has such low accuracy for its conversions.

The tests should hopefully catch any accidental changes in the future so thanks for adding them, Ive kicked off a CI test for them on Linux and will merge if the tests work ok.

[spevans]

@swift-ci test linux

[spevans]

@swift-ci test linux

[spevans]

@ole You'll need to add TestUnitVolume.swift into the file list in TestFoundation/CMakeLists.txt as well - this was the cause of the linux test failure.

[ole]

@spevans Done. Sorry I missed this.

[spevans]

@swift-ci test linux

[spevans]

@swift-ci test linux