Tuesday, January 17, 2012

Hack Intraweb and make it faster!

I'm always looking for a way to make my applications faster. During last couple of days I've been investigating an - already fast - Intraweb application.

1) Intraweb rendering implementation

Lots of Intraweb controls use an utility function to render their HTML. This function called TextToHTML is implemented as a class function of TIWBaseHTMLControl class. This function is used because if you have an IWText component with a caption containing a special character, lets say, an ampersand (&) it must be properly encoded to be shown by the browser. An ampersand declares the beginning of an entity reference (a special character) in HTML, so to be shown in the page it must be replaced by "&" (without quotes). If you have IW sources you can search for TextToHTML and see that it is used in a lot of places, most inside RenderHTML and RenderAsync methods.

2) TextToHTML implementation

Well, TextToHTML is not as fast as it could be. If you imagine that a lot of strings should be processed by TextToHTML before they get to the browser you may improve your application performance if you make that function faster, isn't it? Let's look at TextToHTML code:
class function TIWBaseHTMLControl.TextToHTML(const AText: string; 
  const AConvertEOLs: Boolean; 
  const AConvertSpaces: Boolean): string;
var
  f : integer;
  xIsCallBack: Boolean;
begin
  Result := '';
  xIsCallBack := GGetWebApplicationThreadVar.IsCallBack;
  for f := 1 to Length(AText) do begin
    case AText[f] of
      '<'  : Result := Result + '&lt;';
      '>'  : Result := Result + '&gt;';
      '"'  : Result := Result + '&quot;';
      '''' : Result := Result + '&#39;';
      '&'  : Result := Result + '&amp;';
    else

      {$ifdef UNICODE}
      if (Char(AText[f]) in [#10, #13]) then begin
      {$else}
      if (AnsiChar(AText[f]) in [#10, #13]) then begin
      {$endif}
        if AConvertEOLs then begin
          case AText[f] of
            #10 : Result := Result + '< br >';
            #13 : Result := Result + '';
          end;
        end else begin
          Result := Result + AText[f];
        end;
      end else begin
        if (AText[f] = #32) and AConvertSpaces then begin
          if xIsCallBack then begin
            Result := Result + '&nbsp;';
          end else begin
            Result := Result + ' '
          end;
        end else begin
          Result := Result + AText[f];
        end;
      end;
    end;
  end;
end;
We can see that there is a main loop interating throught all characters of the AText string, concatenating char by char to generate the result. This approach has two drawbacks: (1) Every string concatenation requires a new memory allocation and (2) it is SLOW!

3) A New TextToHTML implementation

We can implement TextToHTML using another approach, used by Delphi's own RTL in functions like HTMLEncode (unit HTTPApp.pas):
class function TIWBaseHTMLControlHack.TextToHTML(const AText: string; 
  const AConvertEOLs: Boolean; 
  const AConvertSpaces: Boolean): string;
var
  Sp, Rp: PChar;
  xIsCallBack: Boolean;
begin
  xIsCallBack := GGetWebApplicationThreadVar.IsCallBack;
  SetLength(Result, Length(AText) * 10);
  Sp := PChar(AText);
  Rp := PChar(Result);
  while Sp^ <> #0 do
  begin
    case Sp^ of
      '&':
        begin
          FormatBuf(Rp^, 5, '&amp;', 5, []);
          Inc(Rp, 4);
        end;
      '<',
        '>':
        begin
          if Sp^ = '<' then
            FormatBuf(Rp^, 4, '&lt;', 4, [])
          else
            FormatBuf(Rp^, 4, '&gt;', 4, []);
          Inc(Rp, 3);
        end;
      '"':
        begin
          FormatBuf(Rp^, 6, '&quot;', 6, []);
          Inc(Rp, 5);
        end;
      '''':
        begin
          FormatBuf(Rp^, 5, '&#39;', 5, []);
          Inc(Rp, 4);
        end;
      '\':
        begin
          FormatBuf(Rp^, 5, '&#92;', 5, []);
          Inc(Rp, 4);
        end;
      #10:
        if AConvertEOLs then
        begin
          FormatBuf(Rp^, 4, '< br >', 4, []);
          Inc(Rp, 3);
        end
        else
          Rp^ := Sp^;
      #13:
        if AConvertEOLs then
        begin
          Dec(Rp);
        end
        else
          Rp^ := Sp^;
      #32:
        if AConvertSpaces then
        begin
          if xIsCallBack then
          begin
            FormatBuf(Rp^, 10, '&amp;nbsp;', 10, []);
            Inc(Rp, 9);
          end else
          begin
            FormatBuf(Rp^, 6, '&nbsp;', 6, []);
            Inc(Rp, 5);
          end;
        end
        else
          Rp^ := Sp^;
    else
      Rp^ := Sp^
    end;
    Inc(Rp);
    Inc(Sp);
  end;
  SetLength(Result, Rp - PChar(Result));
end;
This new implemenation has two main advantages over the former: (1) There is less overhead due memory allocations, and (2) it is FASTER! :)

4) The problem

TextToHTML is a public method of TIWBaseHTMLControl but it is not virtual. Even if it were virtual, we would have to create descendant classes to override this method and it is not acceptable.

5) The solution: Hack it!

Thanks to great code (RtlVclOptimize.pas) created and made available by Andreas Hausladen we can change TIWBaseHTMLControl implementation on the fly, patching the class. I will not enter in detail about Andreas code, but the hack is done replacing, in memory, the TIWBaseHTMLControl.TextToHTML method by another method from some other class. I needed to change RtlVclOptimize.pas a little, putting the declaration of the function "CodeRedirect" in the interface section, so it could be used by my code, outside that unit.
Update: I'm not using Andreas RtlVclOptimize.pas anymore because it has a few incompatibilities with recent Delphi versions. I'm using another unit CodeRedirect.pas, included in the download file.

6) Speed comparison

The use of Andy's RtlVclOptimize unit alone can significantly improve IW applications performance. Moreover, in my tests, my TextToHTML implemenation is more than 4 times faster than the original. Original code: 14.7 seconds versus 3.4 seconds using my code (1,000,000 function calls). Another advantage is less memory allocations for strings (less memory fragmentation and less LOCK+multicore related issues - read more about it here).

7) More Hacks!

Well, not satisfied hacking TextToHTML code, I did the same thing with another TIWBaseHTMLControl method, TextToJSStringLiteral . Finally I've packed all these little things inside a unit. To speed up your IWApplication a bit just declare RtlVclOptimize (from Andreas) and my unit, IntrawebPatch.pas inside your .DPR file and you are done!

8) Finally, the downloads!

You can download my patch unit (IntrawebPatch.pas) and other required files here.

Enjoy!

5 comments:

Anonymous said...

Alenxadre me adicione tenho proposta para voce
powerguidorox@hotmail.com.

fabricioaraujo_rj said...

Alexandre, você contactou a Atozed por causa disso? Seria interessante eles incorporarem isso ao produto, de modo que o hack (me refiro ao código para patch) não seja mais necessário.

Alexandre Machado said...

@fabricoaraujo_rj: Sim, a minha modificação já está sendo considerada pelo pessoal da AtoZed para constar no código do Intraweb, logo acredito que estará disponível. :-)
Abraço

Anonymous said...

There is a problem with the Patched TextoHTML and a TIWCombobox, when the items have an space, it displays erroneous values.

Anonymous said...

IntrawebPatch.pas's dowload link is broken