#85 Error writing big numbers - wrong datatype in

[Karel Rys]

Error in:
function IntToStrWidth(Val: {$ifdef SUPPORT_INT64}Int64{$else}Integer{$endif}; const FieldSize: Integer; const Dest: PAnsiChar; Pad: Boolean; PadChar: AnsiChar): Integer;
var
FloatResult: TFloatResult;
Negative: Boolean;
IntValue: Integer;
Buffer: array[0..{$ifdef SUPPORT_INT64}18{$else}9{$endif}] of AnsiChar;
P: PAnsiChar;

IntValue must be declared as Int64, or probably as
var IntValue: {$ifdef SUPPORT_INT64} Int64 {$else} Integer {$endif};

When using Integer and saving a number 3000000000, you get just garbage written into data field.

[Paul Baker]

What is the type, size and precision of the field? There seems little doubt there is a problem here, but I am having difficulty knowing how to reproduce it.

[Karel Rys]

Is is numeric(12,0). I am attaching an example (field POSLFP).

[Paul Baker]

Here is complete sample code to reproduce the problem.

program Test;

uses
  db,
  dbf;

var
  ADbf: TDbf;
  FieldDef: TFieldDef;
begin
  ADbf := TDbf.Create(nil);
  try
    ADbf.TableName := 'test.dbf';
    FieldDef := ADbf.FieldDefs.Add as TFieldDef;
    FieldDef.Name := 'TEST';
    FieldDef.DataType := ftFloat;
    FieldDef.Size := 12;
    FieldDef.Precision := 0;
    ADbf.CreateTable;
    ADbf.Open;
    ADbf.Insert;
    try
      ADbf['TEST'] := 3000000000;
      ADbf.Post;
    except
      ADbf.Cancel;
      raise;
    end;
  finally
    ADbf.Free;
  end;
end.

If range checking is on, there is a Range Check Error.

If range checking is off, it writes an invalid value to the field, "/.',').'", that cannot be read back. This is because the Val parameter is positive whereas the IntValue parameter, which is always supposed to be positive, is negative.

[Paul Baker]

Fixed in v7.0.2 exactly as Karel proposed.

I don't think I have access to close the ticket.

[Stephane Senecal]

Is this correction is supposed to be in SVN? Beacause it is not.

[Paul Baker]

I see that for the ftLargeInt field type, TDbfFile.SetFieldData() calls IntToStrWidth(), which is declared exactly as above.

This change was first made in the paulenandrew branch with a commit that adds more robust Numeric (N) and Float (F) field value translation. This should also fix bug 85, but did not specifically reference it at the time.
https://sourceforge.net/p/tdbf/code/433

This change was merged into trunk with this commit.
https://sourceforge.net/p/tdbf/code/641

This change was documented as fixing bug 85 in this commit.
https://sourceforge.net/p/tdbf/code/515

Can you reproduce this or a related bug? If so, please provide SVN revision number or release version, steps to reproduce, expected behaviour and actual behaviour.

[Stephane Senecal]

I'm using the released version 7.0.1. You said that it was fixed in 7.0.2, but it is nowhere to be seen.
If you check the latest version of the file dbf_prssupp.pas (https://sourceforge.net/p/tdbf/code/HEAD/tree/trunk/src/dbf_prssupp.pas) at line 356, the variable IntValue is still of type Integer instead of Int64. This cause a range check error at line 367 ou 369 (IntValue := Val) for very large values of Val.

I hope it's just a commit problem. If not, I'll try to make a sample.

[Paul Baker]

You are correct! As you see from the history I mentioned, fixing bug 85 was not the original purpose for my change. I later thought it fixed bug 85 and documented that, but I was wrong. It only changes it to a different problem, a Range Check Error.

I believe I have fixed it in revision 663. Can you verify that, please?
https://sourceforge.net/p/tdbf/code/663

I did not change history.txt, because the existing entry covers it and it was never released like it was.

I remember now that 7.0.2 is not released. There are very few (I think two) people that have rights to this project that are needed to publish a release.

[Stephane Senecal]

It works, thank you.