Faster way to split text in Delphi TStringList

Question

I have an app that needs to do heavy text manipulation in a TStringList. Basically i need to split text by a delimiter ; for instance, if i have a singe line with 1000 chars and this delimiter occurs 3 times in this line, then i need to split it in 3 lines. The delimiter can contain more than one char, it can be a tag like '[test]' for example.

I've wrote two functions to do this task with 2 different approaches, but both are slow in big amounts of text (more then 2mbytes usually).

How can i achieve this goal in a faster way ?

Here are both functions, both receive 2 paramaters : 'lines' which is the original tstringlist and 'q' which is the delimiter.

function splitlines(lines : tstringlist; q: string) : integer;
var
  s, aux, ant : string;
  i,j : integer;
  flag : boolean;
  m2 : tstringlist;
begin
  try
    m2 := tstringlist.create;
    m2.BeginUpdate;
    result := 0;
    for i := 0 to lines.count-1 do
    begin
      s := lines[i];
      for j := 1 to length(s) do
      begin
        flag := lowercase(copy(s,j,length(q))) = lowercase(q);
        if flag then
        begin
          inc(result);
          m2.add(aux);
          aux := s[j];
        end
        else
          aux := aux + s[j];
      end;
      m2.add(aux);
      aux := '';
    end;
    m2.EndUpdate;
    lines.text := m2.text;
  finally
    m2.free;
  end;
end;


function splitLines2(lines : tstringlist; q: string) : integer;
var
  aux, p : string;
  i : integer;
  flag : boolean;
begin
  //maux1 and maux2 are already instanced in the parent class
  try
    maux2.text := lines.text;
    p := '';
    i := 0;
    flag := false;
    maux1.BeginUpdate;
    maux2.BeginUpdate;
    while (pos(lowercase(q),lowercase(maux2.text)) > 0) and (i < 5000) do
    begin
      flag := true;
      aux := p+copy(maux2.text,1,pos(lowercase(q),lowercase(maux2.text))-1);
      maux1.add(aux);
      maux2.text := copy(maux2.text,pos(lowercase(q),lowercase(maux2.text)),length(maux2.text));
      p := copy(maux2.text,1,1);
      maux2.text := copy(maux2.text,2,length(maux2.text));
      inc(i);
    end;
  finally
    result := i;
    maux1.EndUpdate;
    maux2.EndUpdate;
    if flag then
    begin
      maux1.add(p+maux2.text);
      lines.text := maux1.text;
    end;
  end;
end;

Answer 1

I've not tested the speed, but for academic purposes, here's an easy way to split the strings:

myStringList.Text :=
  StringReplace(myStringList.Text, myDelimiter, #13#10, [rfReplaceAll]);
// Use [rfReplaceAll, rfIgnoreCase] if you want to ignore case

When you set the Text property of TStringList, it parses on new lines and splits there, so converting to a string, replacing the delimiter with new lines, then assigning it back to the Text property works.

Answer 2

The problems with your code (at least second approach) are

• You are constantly using lowecase which is slow if called so many times
• If I saw correctly you are copying the whole remaining text back to the original source. This is sure to be extra slow for large strings (eg files)

I have a tokenizer in my library. Its not the fastest or best but it should do (you can get it from Cromis Library, just use the units Cromis.StringUtils and Cromis.Unicode):

type
  TTokens = array of ustring;

  TTextTokenizer = class
  private
    FTokens: TTokens;
    FDelimiters: array of ustring;
  public
    constructor Create;
    procedure Tokenize(const Text: ustring);
    procedure AddDelimiters(const Delimiters: array of ustring);
    property Tokens: TTokens read FTokens;
  end;

{ TTextTokenizer }

procedure TTextTokenizer.AddDelimiters(const Delimiters: array of ustring);
var
  I: Integer;
begin
  if Length(Delimiters) > 0 then
  begin
    SetLength(FDelimiters, Length(Delimiters));

    for I := 0 to Length(Delimiters) - 1 do
      FDelimiters[I] := Delimiters[I];
  end;
end;

constructor TTextTokenizer.Create;
begin
  SetLength(FTokens, 0);
  SetLength(FDelimiters, 0);
end;

procedure TTextTokenizer.Tokenize(const Text: ustring);
var
  I, K: Integer;
  Counter: Integer;
  NewToken: ustring;
  Position: Integer;
  CurrToken: ustring;
begin
  SetLength(FTokens, 100);
  CurrToken := '';
  Counter := 0;

  for I := 1 to Length(Text) do
  begin
    CurrToken := CurrToken + Text[I];

    for K := 0 to Length(FDelimiters) - 1 do
    begin
      Position := Pos(FDelimiters[K], CurrToken);

      if Position > 0 then
      begin
        NewToken := Copy(CurrToken, 1, Position - 1);

        if NewToken <> '' then
        begin
          if Counter > Length(FTokens) then
            SetLength(FTokens, Length(FTokens) * 2);

          FTokens[Counter] := Trim(NewToken);
          Inc(Counter)
        end;

        CurrToken := '';
      end;
    end;
  end;

  if CurrToken <> '' then
  begin
    if Counter > Length(FTokens) then
      SetLength(FTokens, Length(FTokens) * 2);

    FTokens[Counter] := Trim(CurrToken);
    Inc(Counter)
  end;

  SetLength(FTokens, Counter);
end;

Answer 3

How about just using StrTokens from the JCL library

procedure StrTokens(const S: string; const List: TStrings);

It's open source http://sourceforge.net/projects/jcl/

Answer 4

As an additional option, you can use regular expressions. Recent versions of Delphi (XE4 and XE5) come with built in regular expression support; older versions can find a free regex library download (zip file) at Regular-Expressions.info.

For the built-in regex support (uses the generic TArray<string>):

var
  RegexObj: TRegEx;
  SplitArray: TArray<string>;
begin
  SplitArray := nil;
  try
    RegexObj := TRegEx.Create('\[test\]'); // Your sample expression. Replace with q
    SplitArray := RegexObj.Split(Lines, 0);
  except
    on E: ERegularExpressionError do begin
    // Syntax error in the regular expression
    end;
  end;
  // Use SplitArray
end;

For using TPerlRegEx in earlier Delphi versions:

var
  Regex: TPerlRegEx;
  m2: TStringList;
begin
  m2 := TStringList.Create;
  try
    Regex := TPerlRegEx.Create;
    try
      Regex.RegEx := '\[test\]';  //  Using your sample expression - replace with q
      Regex.Options := [];
      Regex.State := [preNotEmpty];
      Regex.Subject := Lines.Text;
      Regex.SplitCapture(m2, 0);
    finally
      Regex.Free;
    end;
    // Work with m2
  finally
    m2.Free;
  end;
end;

(For those unaware, the \ in the sample expression used are because the [] characters are meaningful in regular expressions and need to be escaped to be used in the regular expression text. Typically, they're not required in the text.)