Hi,

[posted this again, as I didn't see it appear on the list]

Is this a bug? The actual value of 234.502 is not greater than either data types (extended or double), but the conversion to string result using the Double is not what I expected.

Strangely enough 234.501 seem to work. 7.502 and 8.502 is another strange one!

procedure TForm1.Button4Click(Sender: TObject);
var
 Result: string;
 e: extended;
 r: double;
begin
 e := 234.502;
 Result := FloatToStrF(e, ffGeneral, 15, 0);
 Memo1.Lines.Add(Result);      { prints 234.502  }
 r := 234.502;
 Result := FloatToStrF(r, ffGeneral, 15, 0);
 Memo1.Lines.Add(Result);  { prints 234.50200000000001 }
 r := 234.501;
 Result := FloatToStrF(r, ffGeneral, 15, 0);
 Memo1.Lines.Add(Result);  { prints 234.501  Why does this work? }
 r := 7.502;
 Result := FloatToStrF(r, ffGeneral, 15, 0);
 Memo1.Lines.Add(Result);  { prints 7.502 }
 r := 8.502;
 Result := FloatToStrF(r, ffGeneral, 15, 0);
 Memo1.Lines.Add(Result);  { prints 8.502000000000001 }
end;

Regards,
- Graeme -

This is all normal. It has to do with the fact that floating point numbers variables are internally base 2 (binary) floating point notation.

So if you assign a certain decimal value, say 0.1234, is converted to the closest binary notation and assigned to the float. Then when you print it is converted back to decimal. Both operations can cause rounding

Any idea how Delphi handles this differently then without rounding errors? This behaviour was picked up in my unit tests I am porting from Delphi to FPC.

Using Double or Real under Delphi worked. Extended was the only type I could use under FPC that gave me the correct results.

Regards,
- Graeme -

Graeme Geldenhuys wrote

Any idea how Delphi handles this differently then without rounding errors? This behaviour was picked up in my unit tests I am porting from Delphi to FPC.

I think the reason is that FloatToStrF() is not overloaded for double and single, so even if you pass a double, it will be written out with extended precision (so you go past the maximum precision of the value and get rounding errors). Please submit a bug.

Jonas