add regression test for NaN handling in errorbar mode.

[dygraphs.git] / datahandler / datahandler.js

/**
 * @license
 * Copyright 2013 David Eberlein (david.eberlein@ch.sauter-bc.com)
 * MIT-licensed (http://opensource.org/licenses/MIT)
 */

/**
 * @fileoverview This file contains the managment of data handlers
 * @author David Eberlein (david.eberlein@ch.sauter-bc.com)
 * 
 * The idea is to define a common, generic data format that works for all data
 * structures supported by dygraphs. To make this possible, the DataHandler
 * interface is introduced. This makes it possible, that dygraph itself can work
 * with the same logic for every data type independent of the actual format and
 * the DataHandler takes care of the data format specific jobs. 
 * DataHandlers are implemented for all data types supported by Dygraphs and
 * return Dygraphs compliant formats.
 * By default the correct DataHandler is chosen based on the options set.
 * Optionally the user may use his own DataHandler (similar to the plugin
 * system).
 * 
 * 
 * The unified data format returend by each handler is defined as so: 
 * series[n][point] = [x,y,(extras)] 
 * 
 * This format contains the common basis that is needed to draw a simple line
 * series extended by optional extras for more complex graphing types. It
 * contains a primitive x value as first array entry, a primitive y value as
 * second array entry and an optional extras object for additional data needed.
 * 
 * x must always be a number.
 * y must always be a number, NaN of type number or null.
 * extras is optional and must be interpreted by the DataHandler. It may be of
 * any type. 
 * 
 * In practice this might look something like this:
 * default: [x, yVal]
 * errorBar / customBar: [x, yVal, [yTopVariance, yBottomVariance] ]
 * 
 */
/*global Dygraph:false */
/*global DygraphLayout:false */

/**
 * 
 * The data handler is responsible for all data specific operations. All of the
 * series data it receives and returns is always in the unified data format.
 * Initially the unified data is created by the extractSeries method
 * @constructor
 */
Dygraph.DataHandler = function () {
};

/**
 * A collection of functions to create and retrieve data handlers.
 * @type {Object.<!Dygraph.DataHandler>}
 */
Dygraph.DataHandlers = {};

(function() {

"use strict";

var handler = Dygraph.DataHandler;

/**
 * X-value array index constant for unified data samples.
 * @const
 * @type {number}
 */
handler.X = 0;

/**
 * Y-value array index constant for unified data samples.
 * @const
 * @type {number}
 */
handler.Y = 1;

/**
 * Extras-value array index constant for unified data samples.
 * @const
 * @type {number}
 */
handler.EXTRAS = 2;

/**
 * Extracts one series from the raw data (a 2D array) into an array of the
 * unified data format.
 * This is where undesirable points (i.e. negative values on log scales and
 * missing values through which we wish to connect lines) are dropped.
 * TODO(danvk): the "missing values" bit above doesn't seem right.
 * 
 * @param {!Array.<Array>} rawData The raw data passed into dygraphs where 
 *     rawData[i] = [x,ySeries1,...,ySeriesN].
 * @param {!number} seriesIndex Index of the series to extract. All other
 *     series should be ignored.
 * @param {!DygraphOptions} options Dygraph options.
 * @return {Array.<[!number,?number,?]>} The series in the unified data format
 *     where series[i] = [x,y,{extras}]. 
 */
handler.prototype.extractSeries = function(rawData, seriesIndex, options) {
};

/**
 * Converts a series to a Point array.
 * 
 * @param {!Array.<[!number,?number,?]>} series The series in the unified 
 *          data format where series[i] = [x,y,{extras}].
 * @param {!string} setName Name of the series.
 * @param {!number} boundaryIdStart Index offset of the first point, equal to the
 *          number of skipped points left of the date window minimum (if any).
 * @return {!Array.<Dygraph.PointType>} List of points for this series.
 */
handler.prototype.seriesToPoints = function(series, setName, boundaryIdStart) {
  // TODO(bhs): these loops are a hot-spot for high-point-count charts. In
  // fact,
  // on chrome+linux, they are 6 times more expensive than iterating through
  // the
  // points and drawing the lines. The brunt of the cost comes from allocating
  // the |point| structures.
  var points = [];
  for ( var i = 0; i < series.length; ++i) {
    var item = series[i];
    var yraw = item[1];
    var yval = yraw === null ? null : handler.parseFloat(yraw);
    var point = {
      x : NaN,
      y : NaN,
      xval : handler.parseFloat(item[0]),
      yval : yval,
      name : setName, // TODO(danvk): is this really necessary?
      idx : i + boundaryIdStart
    };
    points.push(point);
  }
  this.onPointsCreated_(series, points);
  return points;
};

/**
 * Callback called for each series after the series points have been generated
 * which will later be used by the plotters to draw the graph.
 * Here data may be added to the seriesPoints which is needed by the plotters.
 * The indexes of series and points are in sync meaning the original data
 * sample for series[i] is points[i].
 * 
 * @param {!Array.<[!number,?number,?]>} series The series in the unified 
 *     data format where series[i] = [x,y,{extras}].
 * @param {!Array.<Dygraph.PointType>} points The corresponding points passed 
 *     to the plotter.
 * @protected
 */
handler.prototype.onPointsCreated_ = function(series, points) {
};

/**
 * Calculates the rolling average of a data set.
 * 
 * @param {!Array.<[!number,?number,?]>} series The series in the unified 
 *          data format where series[i] = [x,y,{extras}].
 * @param {!number} rollPeriod The number of points over which to average the data
 * @param {!DygraphOptions} options The dygraph options.
 * @return {!Array.<[!number,?number,?]>} the rolled series.
 */
handler.prototype.rollingAverage = function(series, rollPeriod, options) {
};

/**
 * Computes the range of the data series (including confidence intervals).
 * 
 * @param {!Array.<[!number,?number,?]>} series The series in the unified 
 *     data format where series[i] = [x, y, {extras}].
 * @param {!Array.<number>} dateWindow The x-value range to display with 
 *     the format: [min, max].
 * @param {!DygraphOptions} options The dygraph options.
 * @return {Array.<number>} The low and high extremes of the series in the
 *     given window with the format: [low, high].
 */
handler.prototype.getExtremeYValues = function(series, dateWindow, options) {
};

/**
 * Callback called for each series after the layouting data has been
 * calculated before the series is drawn. Here normalized positioning data
 * should be calculated for the extras of each point.
 * 
 * @param {!Array.<Dygraph.PointType>} points The points passed to 
 *          the plotter.
 * @param {!Object} axis The axis on which the series will be plotted.
 * @param {!boolean} logscale Weather or not to use a logscale.
 */
handler.prototype.onLineEvaluated = function(points, axis, logscale) {
};

/**
 * Helper method that computes the y value of a line defined by the points p1
 * and p2 and a given x value.
 * 
 * @param {!Array.<number>} p1 left point ([x,y]).
 * @param {!Array.<number>} p2 right point ([x,y]).
 * @param {!number} xValue The x value to compute the y-intersection for.
 * @return {number} corresponding y value to x on the line defined by p1 and p2.
 * @private
 */
handler.prototype.computeYInterpolation_ = function(p1, p2, xValue) {
  var deltaY = p2[1] - p1[1];
  var deltaX = p2[0] - p1[0];
  var gradient = deltaY / deltaX;
  var growth = (xValue - p1[0]) * gradient;
  return p1[1] + growth;
};

/**
 * Helper method that returns the first and the last index of the given series
 * that lie inside the given dateWindow.
 * 
 * @param {!Array.<[!number,?number,?]>} series The series in the unified 
 *     data format where series[i] = [x,y,{extras}].
 * @param {!Array.<number>} dateWindow The x-value range to display with 
 *     the format: [min,max].
 * @return {!Array.<[!number,?number,?]>} The samples of the series that 
 *     are in the given date window.
 * @private
 */
handler.prototype.getIndexesInWindow_ = function(series, dateWindow) {
  var firstIdx = 0, lastIdx = series.length - 1;
  if (dateWindow) {
    var idx = 0;
    var low = dateWindow[0];
    var high = dateWindow[1];

    // Start from each side of the array to minimize the performance
    // needed.
    while (idx < series.length - 1 && series[idx][0] < low) {
      firstIdx++;
      idx++;
    }
    idx = series.length - 1;
    while (idx > 0 && series[idx][0] > high) {
      lastIdx--;
      idx--;
    }
  }
  if (firstIdx <= lastIdx) {
    return [ firstIdx, lastIdx ];
  } else {
    return [ 0, series.length - 1 ];
  }
};

/**
 * Optimized replacement for parseFloat, which was way too slow when almost
 * all values were type number, with few edge cases, none of which were strings.
 * @param {?number} val
 * @return {number}
 * @protected
 */
handler.parseFloat = function(val) {
  // parseFloat(null) is NaN
  if (val === null) {
    return NaN;
  }

  // Assume it's a number or NaN. If it's something else, I'll be shocked.
  return val;
};

})();