I can reproduce your reported problem using the code below with SS2014, the OLEDB driver and
Seattle and the difference in behaviour when the table has been created with MAX as the column size and a specific number (4096 in my case). I thought I would post this is as an alternative
answer because it not only shows how to investigate this difference systematically
but also identifies why this difference arises (and hence how to avoid it in future).
Please refer to and execute the code below, as written, i.e. with the UseMAX
define
active.
Turning on "Use Debug DCUs" in the the project options before executing the code, immediately
reveals that the described exception occurs in Data.Win.ADODB
at line 4920
Recordset.Fields[TField(FModifiedFields[I]).FieldNo-1].Value := Data
of TCustomADODataSet.InternalPost
and the Debug evaluation window reveals that
Data
at this point is Null
.
Next, notice that
update jdtest set NonNullFieldName = ''
executes in an SSMS2014 Query window without complaint (Command(s) completed successfully.
), so it seems that the
fact that Data
is Null
at line 4920 is what is causing the problem and the next question is "Why?"
Well, the first thing to notice is that the form's caption is displaying ftMemo
Next, comment out the UseMAX
define, recompile and execute. Result: No exception
snd notice that the form's caption is now displaying ftString
.
And that's the reason: Using a specific number for the column size means that
the table metadata retrieved by the RTL causes the client-side Field
to be created
as a TStringField
, whose value you can set by a string assignment statement.
OTOH, when you specify MAX, the resulting client-side Field
is of type ftMemo,
which is one of Delphi's BLOB types and when you assign
string values to an ftMemo field, you are at the mercy of code in Data.DB.Pas , which does all the reading (and writing) to the record buffer using a TBlobStream
. The problem with that is that as far as I can see, after a lot of experiments and tracing through the code, the way a TMemoField uses a BlobStream fails to properly distinguish between updating the field contents to '' and setting the field's value to Null
(as in System.Variants).
In short, whenever you try to set a TMemoField's value to an empty string, what actually happens is that the field's state is set to Null, and this is what causes the exception in the q. AFAICS, this is unavoidable, so no work-around is obvious, to me at any rate.
I have not investigated whether the choice between ftMemo
and ftString
is made by the Delphi RTL code or the MDAC(Ado) layer it sits upon: I would expect it is actually determined by the RecordSet
TAdoQuery uses.
QED. Notice that this systematic approach to debugging has revealed the
problem & cause with very little effort and zero trial and error, which was
what I was trying to suggest in my comments on the q.
Another point is that this problem could be tracked down entirely without
resorting to server-side tools including the SMSS profiler. There wasn't any need to use the profiler to inspect what the client was sending to the server
because there was no reason to suppose that the error returned by the server
was incorrect. That confirms what I said about starting investigation at the client side.
Also, using a table created on the fly using IfDef
ed Sql enabled the problem effectively to be isolated in a single step by simple observation of two runs of the app.
Code
uses [...] TypInfo;
[...]
implementation[...]
const
// The following consts are to create the table and insert a single row
//
// The difference between them is that scSqlSetUp1 specifies
// the size of the NonNullFieldName to 'MAX' whereas scSqlSetUp2 specifies a size of 4096
scSqlSetUp1 =
'CREATE TABLE [dbo].[JDTest]('#13#10
+ ' [ID] [int] NOT NULL primary key,'#13#10
+ ' [NonNullFieldName] VarChar(MAX) NOT NULL'#13#10
+ ') ON [PRIMARY]'#13#10
+ ';'#13#10
+ 'Insert JDTest (ID, [NonNullFieldName]) values (1, ''a'')'#13#10
+ ';'#13#10
+ 'SET ANSI_PADDING OFF'#13#10
+ ';';
scSqlSetUp2 =
'CREATE TABLE [dbo].[JDTest]('#13#10
+ ' [ID] [int] NOT NULL primary key,'#13#10
+ ' [NonNullFieldName] VarChar(4096) NOT NULL'#13#10
+ ') ON [PRIMARY]'#13#10
+ ';'#13#10
+ 'Insert JDTest (ID, [NonNullFieldName]) values (1, ''a'')'#13#10
+ ';'#13#10
+ 'SET ANSI_PADDING OFF'#13#10
+ ';';
scSqlDropTable = 'drop table [dbo].[jdtest]';
procedure TForm1.Test1;
var
AField : TField;
S : String;
begin
// Following creates the table. The define determines the size of the NonNullFieldName
{$define UseMAX}
{$ifdef UseMAX}
S := scSqlSetUp1;
{$else}
S := scSqlSetUp2;
{$endif}
ADOConnection1.Execute(S);
try
ADOQuery1.Open;
try
ADOQuery1.Edit;
// Get explicit reference to the NonNullFieldName
// field to make working with it and investigating it easier
AField := ADOQuery1.FieldByName('NonNullFieldName');
// The following, which requires the `TypInfo` unit in the `USES` list is to find out which exact type
// AField is. Answer: ftMemo, or ftString, depending on UseMAX.
// Of course, we could get this info by inspection in the IDE
// by creating persistent fields
S := GetEnumName(TypeInfo(TFieldType), Ord(AField.DataType));
Caption := S; // Displays `ftMemo` or `ftString`, of course
AField.AsString:= '';
ADOQuery1.Post; //<-- Exception raised while posting
finally
ADOQuery1.Close;
end;
finally
// Tidy up
ADOConnection1.Execute(scSqlDropTable);
end;
end;
procedure TForm1.Button1Click(Sender: TObject);
begin
Test1;
end;
n/vacrchar(MAX)
column can not be set to empty string viaTADODataSet
Edit
/Post
? (I can't test this right now and never used a non-nullnvarchar(MAX)
in my projects) – Insignia