Teafiles和茶馆图表库背后的架构

我目前正在开发一个基于ZedGraph库的趋势解决方案，我正在使用TeaFiles来缓存来自数据库的大量数据。

我不知道茶馆解决方案背后到底是什么样的技术。但我也使用了一种方法来显示来自TeaFile的大量数据中两个日期之间的一组点。

ZedGraph库有一个FilteredPointList对象，它执行自动数据点抽取。它包括一个SetBounds方法，允许您选择要显示的日期范围和要显示的最大点数。通常，它对应于视图的实际宽度。

FilteredPointList(原始源代码)使用两个double数组来包含XY数据。考虑到T是一个包含DateTime和double属性的结构，通过用TeaFile对象替换数组可以很容易地使这个类适应TeaFilePointList。

实现不是最优的，但我开始这样做。稍后我可能会更新这段代码，以包含TeaFile的MemoryMappedFile特性。这样会快得多。

public class TeaFilePointList : IPointList
{
    TeaFile<point> tf;
    private int _maxPts = -1;
    private int _minBoundIndex = -1;
    private int _maxBoundIndex = -1;
    struct point
    {
        public TeaTime.Time x;
        public double y;
    }
    public TeaFilePointList(DateTime[] x, double[] y)
    {
        tf = TeaFile<point>.Create(Path.GetRandomFileName() + ".tea");
        for (var i = 0; i < x.Length; i++)
            tf.Write(new point() { x = x[i], y = y[i] });
    }
    public void SetBounds(double min, double max, int maxPts)
    {
        _maxPts = maxPts;
        // find the index of the start and end of the bounded range
        var xmin = (DateTime)new XDate(min);
        var xmax = (DateTime)new XDate(max);
        int first = tf.BinarySearch(xmin, item => (DateTime)item.x);
        int last = tf.BinarySearch(xmax, item => (DateTime)item.x);
        // Make sure the bounded indices are legitimate
        // if BinarySearch() doesn't find the value, it returns the bitwise
        // complement of the index of the 1st element larger than the sought value
        if (first < 0)
        {
            if (first == -1)
                first = 0;
            else
                first = ~(first + 1);
        }
        if (last < 0)
            last = ~last;
        _minBoundIndex = first;
        _maxBoundIndex = last;
    }
    public int Count
    {
        get
        {
            int arraySize = (int)tf.Count;
            // Is the filter active?
            if (_minBoundIndex >= 0 && _maxBoundIndex >= 0 && _maxPts > 0)
            {
                // get the number of points within the filter bounds
                int boundSize = _maxBoundIndex - _minBoundIndex + 1;
                // limit the point count to the filter bounds
                if (boundSize < arraySize)
                    arraySize = boundSize;
                // limit the point count to the declared max points
                if (arraySize > _maxPts)
                    arraySize = _maxPts;
            }
            return arraySize;
        }
    }
    public PointPair this[int index]
    {
        get
        {
            if (_minBoundIndex >= 0 && _maxBoundIndex >= 0 && _maxPts >= 0)
            {
                // get number of points in bounded range
                int nPts = _maxBoundIndex - _minBoundIndex + 1;
                if (nPts > _maxPts)
                {
                    // if we're skipping points, then calculate the new index
                    index = _minBoundIndex + (int)((double)index * (double)nPts / (double)_maxPts);
                }
                else
                {
                    // otherwise, index is just offset by the start of the bounded range
                    index += _minBoundIndex;
                }
            }
            double xVal, yVal;
            if (index >= 0 && index < tf.Count)
                xVal = new XDate(tf.Items[index].x);
            else
                xVal = PointPair.Missing;
            if (index >= 0 && index < tf.Count)
                yVal = tf.Items[index].y;
            else
                yVal = PointPair.Missing;
            return new PointPair(xVal, yVal, PointPair.Missing, null);
        }
    }
    public object Clone()
    {
        throw new NotImplementedException(); // I'm lazy...
    }
    public void Close()
    {
        tf.Close();
        tf.Dispose();
        File.Delete(tf.Name);
    }
}

最难的部分是实现TeaFile的BinarySearch，以便使用DateTime快速搜索记录。我看了看阵列。通过使用反编译器实现BinarySearch，并且我在下面编写了扩展:

public static int BinarySearch<T, U>(this TeaFile<T> tf, U target, Func<T, U> indexer) where T : struct
{
    var lo = 0;
    var hi = (int)tf.Count - 1;
    var comp = Comparer<U>.Default;
    while(lo <= hi)
    {
        var median = lo + (hi - lo >> 1);
        var num = comp.Compare(indexer(tf.Items[median]), target);
        if (num == 0)
            return median;
        if (num < 0)
            lo = median + 1;
        else
            hi = median - 1;
    }
    return ~lo;
}

如果ZedGraph不适合你的需要，至少你得到了这个想法。FilteredPointList类中使用的抽取算法非常好，可以以另一种方式进行调整以满足您的需要。