单元测试 - 有时有效，有时无效

Question

我在运行下面针对后面的代码进行了单元测试。这个测试有时通过，有时失败。不知道为什么并且犹豫是否要从根本上改变事情，因为它是一个公式，有时会通过......我认为它可能与 double 类型的精度有关？不确定。想法？

[TestMethod]
public void CircleFromCircumference()
{
    var random = new Random();
    var circumference = random.NextDouble();
    var circle = new Circle("My circle", circumference, Circle.CircleDimensions.Circumference);

    var var1 = circumference - circle.Circumference;
    var var2 = circumference - 2 * Math.PI * circle.Radius;
    var var3 = circumference - Math.PI * circle.Diameter;
    var var4 = Math.Pow(circumference / (2 * Math.PI), 2) * Math.PI - circle.Area;

    Assert.IsTrue(
        circumference - circle.Circumference <= 0 //circumference
        && circumference - 2 * Math.PI * circle.Radius <= 0 //radius
        && circumference - Math.PI * circle.Diameter <= 0 //diameter
        && Math.Pow(circumference / (2 * Math.PI), 2) * Math.PI - circle.Area <= 0 //area
        && string.IsNullOrEmpty(circle.ShapeException));
}



using System;
using System.Runtime.Serialization;

namespace Shapes
{
    [DataContract]
    public class Circle : Shape
    {
        [DataMember] public double Radius { get; set; }
        [DataMember] public double Diameter { get; set; }
        [DataMember] public double Circumference { get; set; }

        /// <summary>
        /// The name of the value you are sending. Radius is the default
        /// </summary>
        public enum CircleDimensions
        {
            Circumference = 1,
            Area = 2,
            Diameter = 3
        }
        /// <summary>
        /// 
        /// </summary>
        /// <param name="circleName">The name of your circle</param>
        /// <param name="dimension">The value of the dimension you are providing</param>
        /// <param name="circleDimensions">The name of the value you are providing. Radius is default</param>
        public Circle(string circleName, double dimension = 0, CircleDimensions circleDimensions = 0)
        {
            this.ShapeName = circleName;
            if (dimension <= 0)
            {
                this.ShapeException = "Parameters must be greater than zero";
                return;
            }

            switch (circleDimensions)
            {
                case CircleDimensions.Circumference:
                    //radius from Circumference
                    this.Circumference = dimension;
                    this.Radius = this.RadiusFromCircumference(dimension);
                    this.Area = this.CalculateArea(this.Radius);
                    this.Diameter = this.CalculateDiameter(this.Radius);
                    break;
                case CircleDimensions.Area:
                    //radius from Area
                    break;
                case CircleDimensions.Diameter:
                    //radius from diameter
                    break;
                default: //calculate from radius
                    this.Radius = dimension;
                    this.Diameter = this.CalculateDiameter(dimension);
                    this.Circumference = this.CalculateCircumference(dimension);
                    this.Area = this.CalculateArea(dimension);
                    break;
            }
        }

        private double RadiusFromCircumference(double dimension) => dimension / (2 * Math.PI);

        private double CalculateCircumference(double dimension) => 2 * Math.PI * dimension;
        private double CalculateDiameter(double dimension) => 2 * dimension;
        private double CalculateArea(double dimension) =>
            Math.PI * (Math.Pow(dimension, 2));
    }
}

Answer 1

不一致与精度本身无关，它更多地与浮点表示的工作方式有关。例如，如果您这样写：

for (float f = 0.0f; f != 1.0f; f+=0.1f)
{
   Console.WriteLine(f);
}

永远不会退出。因为 0.1 没有二进制形式的精确表示。 (https://www.exploringbinary.com/why-0-point-1-does-not-exist-in-floating-point/). I also recommend reading (https://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.html)

回到手头的问题，在你的代码中，你正在使用这个获得半径：

dimension / (2 * Math.PI); //passed in dimension is the Circumference, returns radius

然后在你的测试中你断言：

circumference - 2 * Math.PI * circle.Radius <= 0

不能保证用同一个浮点数除以然后乘以得到原始浮点数。

因此，断言这一点通常不是一个好主意。最常见的测试 "almost equality" 的方法是测试相等性 "within limits"。在您的情况下，您所要做的就是在测试中定义一个您认为 "acceptable" 大于或等于 double.Epsilon 的足够小的 epsilon。

var epsilon = double.Epsilon;
Assert.IsTrue(
    Math.Abs(circumference - circle.Circumference) <= epsilon //circumference
    && Math.Abs(circumference - 2 * Math.PI * circle.Radius) <= epsilon //radius
    && Math.Abs(circumference - Math.PI * circle.Diameter) <= epsilon //diameter
    && Math.Abs(Math.Pow(circumference / (2 * Math.PI), 2) * Math.PI - circle.Area) <= epsilon //area
    && string.IsNullOrEmpty(circle.ShapeException));

如果相反，您必须保证准确性，一种选择是切换到非浮点类型，如十进制，但预计会影响性能。

单元测试 - 有时有效，有时无效

Unit Test - sometimes works, sometimes not

c#

double

mstest