Hello,

Usually when we work with PChars we do things like this to set the final terminator:

var
  ansistr: ansistring;
begin
  // Copy the originating string taking into account the specified length
  SetLength(ansistr, len+1);
  System.Move(source^, ansistr[1], len);
  ansistr[len+1] := #0;
  SetLength(ansistr, len);

And it works fine. You get a string with correct length and a final #0 terminator.

Now, I simply tryed the same code for a unicodestring, and it corrupts my stack:

procedure Unicode2AnsiMove(source:pwidechar;var dest:ansistring;len:SizeInt);
var
  widestr: unicodestring;
begin
  {$ifdef PASWSTRING_VERBOSE}WriteLn('Unicode2AnsiMove START');{$endif}
  // Copy the originating string taking into account the specified length
  SetLength(widestr, len+1);
  System.Move(source^, widestr[1], len*2);
  PWideChar(@widestr)[len] := #0; // <--- This corrupts the stack
  SetLength(widestr, len);

I already fixed the problem by simply not setting this last widechar to #0, but now I got puzzled: Any ideas why this doesn't work?

It crashes only a bit after this procedure, but I found that this was the statement which caused the problem.

Why don't you use the same statement as for the AnsiString version?

E.g.

widestr[len+1] := #0;

Regards,
Sven

On Wed, October 19, 2011 09:13, Felipe Monteiro de Carvalho wrote:

Hi,

Sorry for a silly question, but why the difference between the two implementations (i.e. why not addressing the WideStr element directly without the typecast)? Moreover, even if using the PWideChar, I believe that the correct version would be without the '@' (i.e.: PWideChar(widestr)[len] := #0), wouldn't it?

Tomas

Felipe Monteiro de Carvalho hat am 19.
Oktober 2011 um 09:13 geschrieben:

I never saw such code for ansistring. Only for shortstring.  SetLength(ansistring) already allocates memory for the header+len+1, creates the header and sets the len+1 character to #0. 
Simply do:
  SetLength(ansistr, len);
  if len>1 then
    System.Move(source^, ansistr[1], len);
 

This would corrupt the stack for ansistring too. This would only work for shortstring.
widestr is a unicodestring is a pointer to widechar.    
  PWideChar(widestr)[len] := #0; 
 
 

Mattias

2011/10/19 Tomas Hajny :

Sorry for a silly question, but why the difference between the two implementations (i.e. why not addressing the WideStr element directly without the typecast)?

Actually it should be a pointer operation in the ansistring version too, because setting chars in a string calls UniqueString, which is undesirable and inneficient.

Moreover, even if using the PWideChar, I believe that the correct version would be without the '@' (i.e.: PWideChar(widestr)[len] := #0), wouldn't it?

Good idea! I think you found the problem.

Mattias Gaertner wrote

I never saw such code for ansistring. Only for shortstring.

Indeed, I copied some code from the RTL and it seams that I inadvertedly copied code for shortstring instead of ansistring =o

I fixed this in rev 32977 in the Pascal Widestring Manager