string.split()读取制表符分隔文件时出现"内存不足异常"

Question

我在我的C#代码中使用string.split()来读取制表符分隔文件.我正面临着代码示例中提到的"OutOfMemory异常".

在这里,我想知道为什么文件大小为16 MB会出现问题？

这是正确的方法还是没有？

using (StreamReader reader = new StreamReader(_path))
{
  //...........Load the first line of the file................
  string headerLine = reader.ReadLine();

  MeterDataIPValueList objMeterDataList = new MeterDataIPValueList();
  string[] seperator = new string[1];   //used to sepreate lines of file

  seperator[0] = "\r\n";
  //.............Load Records of file into string array and remove all empty lines of file.................
  string[] line = reader.ReadToEnd().Split(seperator, StringSplitOptions.RemoveEmptyEntries);
  int noOfLines = line.Count();
  if (noOfLines == 0)
  {
    mFileValidationErrors.Append(ConstMsgStrings.headerOnly + Environment.NewLine);
  }
  //...............If file contains records also with header line..............
  else
  {
    string[] headers = headerLine.Split('\t');
    int noOfColumns = headers.Count();

    //.........Create table structure.............
    objValidateRecordsTable.Columns.Add("SerialNo");
    objValidateRecordsTable.Columns.Add("SurveyDate");
    objValidateRecordsTable.Columns.Add("Interval");
    objValidateRecordsTable.Columns.Add("Status");
    objValidateRecordsTable.Columns.Add("Consumption");

    //........Fill objValidateRecordsTable table by string array contents ............

    int recordNumber;  // used for log
    #region ..............Fill objValidateRecordsTable.....................
    seperator[0] = "\t";
    for (int lineNo = 0; lineNo < noOfLines; lineNo++)
    {
      recordNumber = lineNo + 1;
      **string[] recordFields = line[lineNo].Split(seperator, StringSplitOptions.RemoveEmptyEntries);** // Showing me error when we  split columns
      if (recordFields.Count() == noOfColumns)
      {
        //Do processing
      }

Answer 1

Split的实现很差,并且在应用于大字符串时会出现严重的性能问题.有关分割功能的内存要求的详细信息,请参阅此文章:

当你对包含1355049个逗号分隔的字符串的字符串进行拆分时会发生什么,每个字符串包含16个字符,总字符长度为25745930？

1. 字符串对象的指针数组:连续虚拟地址空间4(地址指针)*1355049 = 5420196(数组大小)+ 16(用于簿记)= 5420212.

2. 1355049个字符串的非连续虚拟地址空间,每个字符串54个字节.这并不意味着所有这130万个字符串都会分散在整个堆中,但它们不会在LOH上分配.GC将在Gen0堆上的串上分配它们.

3. Split.Function将创建大小为25745930的System.Int32 []的内部数组,消耗(102983736字节)~98MB的LOH,这是非常昂贵的L.

Answer 2

尽量不要将整个文件读入数组"reader.ReadToEnd()"直接读取文件.

  using (StreamReader sr = new StreamReader(this._path))
        {
            string line = "";
            while(( line= sr.ReadLine()) != null)
            {
                string[] cells = line.Split(new string[] { "\t" }, StringSplitOptions.None);
                if (cells.Length > 0)
                {

                }
            }
        }

Answer 3

我用我自己的。它已经通过 10 个单元测试进行了测试。

public static class StringExtensions
{

    // the string.Split() method from .NET tend to run out of memory on 80 Mb strings. 
    // this has been reported several places online. 
    // This version is fast and memory efficient and return no empty lines. 
    public static List<string> LowMemSplit(this string s, string seperator)
    {
        List<string> list = new List<string>();
        int lastPos = 0;
        int pos = s.IndexOf(seperator);
        while (pos > -1)
        {
            while(pos == lastPos)
            {
                lastPos += seperator.Length;
                pos = s.IndexOf(seperator, lastPos);
                if (pos == -1)
                    return list;
            }

            string tmp = s.Substring(lastPos, pos - lastPos);
            if(tmp.Trim().Length > 0)
                list.Add(tmp);
            lastPos = pos + seperator.Length;
            pos = s.IndexOf(seperator, lastPos);
        }

        if (lastPos < s.Length)
        {
            string tmp = s.Substring(lastPos, s.Length - lastPos);
            if (tmp.Trim().Length > 0)
                list.Add(tmp);
        }

        return list;
    }
}