学步园

http://www.codeproject.com/vb/net/SimulatingRecordsets.asp

Introduction

As any .NET developer knows, the ADO.NET approach to data access is substantially different from the ADODB predecessor. First of all, because it is disconnected and mostly based on the DataSet concept (that involves a client-side data caching), while ADODB was normally used as a connected data access paradigm (with the exception of the so-called "disconnected recordsets"). The only way to use ADO.NET in a connected fashion is using objects like DataReader, Command, and Transaction, that are not so comfortable if you need to scroll a result set making updates to some data based on a row-oriented logic. This was a very common task when working with ADODB, and a lot of programmers coming from a Visual Studio 6.0 experience will miss the Recordset concept: being oriented to disconnected scenarios, ADO.NET currently doesn't support features like server-side cursors, and so it doesn't expose objects similar to the ADODB.Recordset that was very useful to implement row-based logics. Anyone prevents you from continuing to use the ADODB objects while programming on .NET, but if you want to avoid the COM interoperability overhead, this is not the right way.

In this article, I propose a class that simulates the behavior of an ADODB.Recordset on a Microsoft SQL Server 2000 database through the use of ADO.NET "connected objects" (Connection, Command, DataReader,...) and of server-side cursors directly implemented in T-SQL. The proposed class is developed for SQL Server 2000, but can be easily modified to work with other RDBMSs.

How the code works

The class I wrote is named Recordset and it tries to simulate the ADODB.Recordset in its main functionalities. Then, it exposes methods like Open(), Close(), MoveNext(), MovePrevious(), MoveFirst(), MoveLast(), Update() and so on (even if it doesn't currently expose an AddNew() method). To support navigation and random access to rows of a result set without caching data on the client, you need to use a scrollable server-side cursor; this cursor has to be and remain open for all the duration of the connected updates. That's why, behind the scenes of the Recordset.Open() method, a connection is open and a T-SQL cursor is created, based on a given SELECT expression:

cnn = <span class="vb-keyword">New</span> SqlConnection(mConnectionString)
cmd = cnn.CreateCommand()
cnn.Open()
...
cmd.CommandText = <span class="vb-string">"DECLARE crsr SCROLL CURSOR FOR "</span> &amp; mSelectString
cmd.ExecuteNonQuery()
cmd.CommandText = <span class="vb-string">"OPEN crsr"</span>
cmd.ExecuteNonQuery()

The various movements inside the Recordset have their counterparts in the server-side T-SQL cursor, so it's not difficult to implement for the Recordset class the following methods:

Method	T-SQL equivalent
MoveNext()	FETCH NEXT FROM crsr
MovePrevious()	FETCH PRIOR FROM crsr
MoveFirst()	FETCH FIRST FROM crsr
MoveLast()	FETCH LAST FROM crsr
MoveAbsolute(n)	FETCH ABSOLUTE n FROM crsr
MoveRelative(n)	FETCH RELATIVE n FROM crsr

For the Recordset.Update() method, if we suppose the cursor being based on a single-table SELECT statement, we can think to code it as a T-SQL statement like the following:

<span class="vb-function">UPDATE</span> TABLE_NAME
<span class="vb-function">SET</span> field1=value1, field2=value2,...
<span class="vb-function">WHERE</span> CURRENT OF crsr

In the same way (under the same single-table SELECT statement restriction), also the Recordset.Delete() method can be coded as:

<span class="vb-function">DELETE</span> TABLE_NAME <span class="vb-function">WHERE</span> CURRENT OF crsr

Finally, the Recordset.Close() method has simply to execute some cleanup code (on the server-side cursor and on the open connection):

cmd.CommandText = <span class="vb-string">"CLOSE crsr"</span>
cmd.ExecuteNonQuery()
cmd.CommandText = <span class="vb-string">"DEALLOCATE crsr"</span>
cmd.ExecuteNonQuery()
cmd.Dispose()
cnn.Close()
cnn.Dispose()