delphi - 如何使用Delphi读取文本文件中的最后一行

Question

我需要读取一些非常大的文本文件中的最后一行(以从数据中获取时间戳)。 TStringlist 是一种简单的方法，但它会返回内存不足错误。我正在尝试使用 seek 和 blockread，但是缓冲区中的字符都是无意义的。这与unicode有关吗？

    Function TForm1.ReadLastLine2(FileName: String): String;
    var
      FileHandle: File;
      s,line: string;
      ok: 0..1;
      Buf: array[1..8] of Char;
      k: longword;
      i,ReadCount: integer;
    begin
      AssignFile (FileHandle,FileName);
      Reset (FileHandle);           // or for binary files: Reset (FileHandle,1);
      ok := 0;
      k := FileSize (FileHandle);
      Seek (FileHandle, k-1);
      s := '';
      while ok<>1 do begin
        BlockRead (FileHandle, buf, SizeOf(Buf)-1, ReadCount);  //BlockRead ( var FileHandle : File; var Buffer; RecordCount : Integer {; var RecordsRead : Integer} ) ;
        if ord (buf[1]) <>13 then         //Arg to integer
          s := s + buf[1]
        else
          ok := ok + 1;
        k := k-1;
        seek (FileHandle,k);
      end;
      CloseFile (FileHandle);

      // Reverse the order in the line read
      setlength (line,length(s));
      for i:=1 to length(s) do
        line[length(s) - i+1 ] := s[i];
      Result := Line;
    end;

基于 www.delphipages.com/forum/showthread.php?t=102965

测试文件是我在 excel 中创建的一个简单的 CSV(这不是我最终需要阅读的 100MB)。

    a,b,c,d,e,f,g,h,i,j,blank
    A,B,C,D,E,F,G,H,I,J,blank
    1,2,3,4,5,6,7,8,9,0,blank
    Mary,had,a,little,lamb,His,fleece,was,white,as,snow
    And,everywhere,that,Mary,went,The,lamb,was,sure,to,go

Answer 1

您确实必须从尾部到头部以大块读取文件。由于它太大，不适合内存 - 然后从头到尾逐行读取它会非常慢。使用 ReadLn - 慢两倍。

您还必须准备好最后一行可能以 EOL 结束，也可能不结束。

就我个人而言，我还会考虑三种可能的 EOL 序列:

• CR/LF aka #13#10=^M^J - DOS/Windows 风格
• 没有 LF 的 CR - 只是 #13=^M - 经典 MacOS 文件
• 没有 CR 的 LF - 只是 #10=^J - UNIX 风格，包括 MacOS 版本 10

如果您确定您的 CSV 文件只会由 native Windows 程序生成，那么可以安全地假设使用了完整的 CR/LF。但如果可以有其他 Java 程序、非 Windows 平台、移动程序——我就不太确定了。当然，没有 LF 的纯 CR 是所有情况中可能性最小的。

uses System.IOUtils, System.Math, System.Classes;

type FileChar = AnsiChar; FileString = AnsiString; // for non-Unicode files
// type FileChar = WideChar; FileString = UnicodeString;// for UTF16 and UCS-2 files
const FileCharSize = SizeOf(FileChar);
// somewhere later in the code add: Assert(FileCharSize = SizeOf(FileString[1]);

function ReadLastLine(const FileName: String): FileString; overload; forward;

const PageSize = 4*1024; 
// the minimal read atom of most modern HDD and the memory allocation atom of Win32
// since the chances your file would have lines longer than 4Kb are very small - I would not increase it to several atoms.

function ReadLastLine(const Lines: TStringDynArray): FileString; overload;
var i: integer;
begin
  Result := '';
  i := High(Lines);
  if i < Low(Lines) then exit; // empty array - empty file

  Result := Lines[i];
  if Result > '' then exit; // we got the line

  Dec(i); // skip the empty ghost line, in case last line was CRLF-terminated
  if i < Low(Lines) then exit; // that ghost was the only line in the empty file
  Result := Lines[i];
end;

// scan for EOLs in not-yet-scanned part
function FindLastLine(buffer: TArray<FileChar>; const OldRead : Integer; 
     const LastChunk: Boolean; out Line: FileString): boolean;
var i, tailCRLF: integer; c: FileChar;
begin
  Result := False;
  if Length(Buffer) = 0 then exit;

  i := High(Buffer);    
  tailCRLF := 0; // test for trailing CR/LF
  if Buffer[i] = ^J then begin // LF - single, or after CR
     Dec(i);
     Inc(tailCRLF);
  end;
  if (i >= Low(Buffer)) and (Buffer[i] = ^M) then begin // CR, alone or before LF
     Inc(tailCRLF);
  end;

  i := High(Buffer) - Max(OldRead, tailCRLF);
  if i - Low(Buffer) < 0 then exit; // no new data to read - results would be like before

  if OldRead > 0 then Inc(i); // the CR/LF pair could be sliced between new and previous buffer - so need to start a bit earlier

  for i := i downto Low(Buffer) do begin
      c := Buffer[i];
      if (c=^J) or (c=^M) then begin // found EOL
         SetString( Line, @Buffer[i+1], High(Buffer) - tailCRLF - i);
         exit(True); 
      end;
  end;  

  // we did not find non-terminating EOL in the buffer (except maybe trailing),
  // now we should ask for more file content, if there is still left any
  // or take the entire file (without trailing EOL if any)

  if LastChunk then begin
     SetString( Line, @Buffer[ Low(Buffer) ], Length(Buffer) - tailCRLF);
     Result := true;
  end;
end;


function ReadLastLine(const FileName: String): FileString; overload;
var Buffer, tmp: TArray<FileChar>; 
    // dynamic arrays - eases memory management and protect from stack corruption
    FS: TFileStream; FSize, NewPos: Int64; 
    OldRead, NewLen : Integer; EndOfFile: boolean;
begin
  Result := '';
  FS := TFile.OpenRead(FileName);
  try
    FSize := FS.Size;
    if FSize <= PageSize then begin // small file, we can be lazy!
       FreeAndNil(FS);  // free the handle and avoid double-free in finally
       Result := ReadLastLine( TFile.ReadAllLines( FileName, TEncoding.ANSI )); 
          // or TEncoding.UTF16
          // warning - TFIle is not share-aware, if the file is being written to by another app
       exit;
    end;

    SetLength( Buffer, PageSize div FileCharSize);
    OldRead := 0;
    repeat
      NewPos := FSize - Length(Buffer)*FileCharSize;
      EndOfFile := NewPos <= 0;
      if NewPos < 0 then NewPos := 0; 
      FS.Position := NewPos;

      FS.ReadBuffer( Buffer[Low(Buffer)], (Length(Buffer) - OldRead)*FileCharSize);

      if FindLastLine(Buffer, OldRead, EndOfFile, Result) then 
         exit; // done !

      tmp := Buffer; Buffer := nil; // flip-flop: preparing to broaden our mouth

      OldRead := Length(tmp); // need not to re-scan the tail again and again when expanding our scanning range
      NewLen := Min( 2*Length(tmp), FSize div FileCharSize );

      SetLength(Buffer, NewLen); // this may trigger EOutOfMemory...
      Move( tmp[Low(tmp)], Buffer[High(Buffer)-OldRead+1], OldRead*FileCharSize);
      tmp := nil; // free old buffer
    until EndOfFile;
  finally
    FS.Free;
  end;
end;

附言。请注意一个额外的特殊情况 - 如果您使用 Unicode 字符(两个字节的字符)并给出奇数长度的文件(3 个字节、5 个字节等)-您将永远无法扫描起始的单个字节(半宽字符).也许你应该在那里添加额外的保护，比如 Assert( 0 = FS.Size mod FileCharSize)

PPS。根据经验，您最好将这些函数保留在表单类之外，因为为什么要混合它们？一般来说，您应该将关注点分成小块。读取文件与用户交互无关 - 因此最好将其卸载到额外的 UNIT。然后，您将能够在主线程或多线程应用程序中以一种形式或 10 种形式使用该单元的函数。就像乐高积木零件一样 - 它们体积小且独立，为您提供了灵 active 。

PPPS。这里的另一种方法是使用内存映射文件。 Google for Delphi 的 MMF implementations 和关于 MMF 方法的好处和问题的文章。我个人认为重写上面的代码以使用 MMF 会大大简化它，删除几个“特例”和麻烦的内存复制触发器。 OTOH 它会要求你对指针算法非常严格。

• https://en.wikipedia.org/wiki/Memory-mapped_file
• https://msdn.microsoft.com/en-us/library/ms810613.aspx
• http://torry.net/quicksearchd.php?String=memory+map&Title=No