What is the point of DBNull?

In .NET there is the null reference, which is used everywhere to denote that an object reference is empty, and then there is the DBNull, which is used by database drivers (and few others) to denote... pretty much the same thing. Naturally, this creates a lot of confusion and conversion routines have to be churned out, etc.

So why did the original .NET authors decide to make this? To me it makes no sense. Their documentation makes no sense either:

The DBNull class represents a nonexistent value. In a database, for example, a column in a row of a table might not contain any data whatsoever. That is, the column is considered to not exist at all instead of merely not having a value. A DBNull object represents the nonexistent column. Additionally, COM interop uses the DBNull class to distinguish between a VT_NULL variant, which indicates a nonexistent value, and a VT_EMPTY variant, which indicates an unspecified value.

What's this crap about a "column not existing"? A column exists, it just doesn't have a value for the particular row. If it didn't exist, I'd get an exception trying to access the specific cell, not a DBNull! I can understand the need to differentiate between VT_NULL and VT_EMPTY, but then why not make a COMEmpty class instead? That would be a much neater fit in the whole .NET framework.

Am I missing something? Can anyone shed some light why DBNull was invented and what problems it helps to solve?

Answers


The point is that in certain situations there is a difference between a database value being null and a .NET Null.

For example. If you using ExecuteScalar (which returns the first column of the first row in the result set) and you get a null back that means that the SQL executed did not return any values. If you get DBNull back it means a value was returned by the SQL and it was NULL. You need to be able to tell the difference.


I'm going to disagree with the trend here. I'll go on record:

I do not agree that DBNull serves any useful purpose; it adds unnecessary confusion, while contributing virtually no value.

The argument is often put forward that null is an invalid reference, and that DBNull is a null object pattern; neither is true. For example:

int? x = null;

this is not an "invalid reference"; it is a null value. Indeed null means whatever you want it to mean, and frankly I have absolutely no problem working with values that may be null (indeed, even in SQL we need to correctly work with null - nothing changes here). Equally, the "null object pattern" only makes sense if you are actually treating it as an object in OOP terms, but if we have a value that can be "our value, or a DBNull" then it must be object, so we can't be doing anything useful with it.

There are so many bad things with DBNull:

  • it forces you to work with object, since only object can hold DBNull or another value
  • there is no real difference between "could be a value or DBNull" vs "could be a value or null"
  • the argument that it stems from 1.1 (pre-nullable-types) is meaningless; we could use null perfectly well in 1.1
  • most APIs have "is it null?" methods, for example DBDataReader.IsDBNull or DataRow.IsNull - neither of which actually require DBNull to exist in terms of the API
  • DBNull fails in terms of null-coalescing; value ?? defaultValue doesn't work if the value is DBNull
  • DBNull.Value can't be used in optional parameters, since it isn't a constant
  • the runtime semantics of DBNull are identical to the semantics of null; in particular, DBNull actually equals DBNull - so it does not do the job of representing the SQL semantic
  • it often forces value-type values to be boxed since it over-uses object
  • if you need to test for DBNull, you might as well have tested for just null
  • it causes huge problems for things like command-parameters, with a very silly behaviour that if a parameter has a null value it isn't sent... well, here's an idea: if you don't want a parameter sent - don't add it to the parameters collection
  • every ORM I can think of works perfectly well without any need or use of DBNull, except as an extra nuisance when talking to the ADO.NET code

The only even remotely compelling argument I've ever seen to justify the existence of such a value is in DataTable, when passing in values to create a new row; a null means "use the default", a DBNull is explicitly a null - frankly this API could have had a specific treatment for this case - an imaginary DataRow.DefaultValue for example would be much better than introducing a DBNull.Value that infects vast swathes of code for no reason.

Equally, the ExecuteScalar scenario is... tenuous at best; if you are executing a scalar method, you expect a result. In the scenario where there are no rows, returning null doesn't seem too terrible. If you absolutely need to disambiguate between "no rows" and "one single null returned", there's the reader API.

This ship has sailed long ago, and it is far far too late to fix it. But! Please do not think that everyone agrees that this is an "obvious" thing. Many developers do not see value in this odd wrinkle on the BCL.

I actually wonder if all of this stems from two things:

  • having to use the word Nothing instead of something involving "null" in VB
  • being able to us the if(value is DBNull) syntax which "looks just like SQL", rather than the oh-so-tricky if(value==null)

Summary:

Having 3 options (null, DBNull, or an actual value) is only useful if there is a genuine example where you need to disambiguate between 3 different cases. I have yet to see an occasion where I need to represent two different "null" states, so DBNull is entirely redundant given that null already exists and has much better language and runtime support.


DbNull represents a box with no contents; null indicates the non-existence of the box.


You use DBNull for missing data. Null in the .NET language means that there is no pointer for an object/variable.

DBNull missing data: http://msdn.microsoft.com/en-us/library/system.dbnull.value.aspx

The effects of missing data on statistics:

http://en.wikipedia.org/wiki/Missing_values


There are some differences between a CLR null and a DBNull. First, null in relational databases has different "equals" semantics: null is not equal to null. CLR null IS equal to null.

But I suspect the main reason is to do with the way parameter default values work in SQL server and the implementation of the provider.

To see the difference, create a procedure with a parameter that has a default value:

CREATE PROC [Echo] @s varchar(MAX) = 'hello'
AS
    SELECT @s [Echo]

Well-structured DAL code should separate command creation from use (to enable using the same command many times, for example to invoke a stored procedure many times efficiently). Write a method that returns a SqlCommand representing the above procedure:

SqlCommand GetEchoProc()
{
    var cmd = new SqlCommand("Echo");
    cmd.Parameters.Add("@s", SqlDbType.VarChar);
    return cmd;
}

If you now invoke the command without setting the @s parameter, or set its value to (CLR) null, it will use the default value 'hello'. If on the other hand you set the parameter value to DBNull.Value, it will use that and echo DbNull.Value.

Since there's two different results using CLR null or database null as parameter value, you can't represent both cases with only one of them. If CLR null was to be the only one, it'd have to work the way DBNull.Value does today. One way to indicate to the provider "I want to use the default value" could then be to not declare the parameter at all (a parameter with a default value of course makes sense to describe as an "optional parameter"), but in a scenario where the command object is cached and reused this does lead to removing and re-adding the parameter.

I'm not sure if I think DBNull was a good idea or not, but a lot of people are unaware of the things I've mentioned here, so I figured it worth mentioning.


Need Your Help

How to format a number from 1123456789 to 1,123,456,789 in C?

c formatting numbers

How can I in C language format a number from 1123456789 to 1,123,456,789?

exception: can't convert from BSON type EOO to Date

mongodb date mongoose aggregation-framework

I am getting an issue for running the following aggregate query: