Question

When returning data from a DataReader I would typically use the ordinal reference on the DataReader to grab the relevant column:

if (dr.HasRows)         
   Console.WriteLine(dr[0].ToString());

or

if (dr.HasRows)         
   Console.WriteLine(dr.GetString(0));

or

if (dr.HasRows)         
   Console.WriteLine((string)dr[0]);

I have always done this because I was advised at an early stage that using dr["ColumnName"] or a more elegant way of indexing causes a performance hit.

However, while all references to data entities are becoming increasingly strongly-typed I feel more uncomfortable with this. I’m also aware that the above does not check for DBNull.

What is the most robust way to return data from a DataReader?

Answer 1

It is possible to argue both sides in this situation. As already pointed out by others, using the name is more readable and will not break if someone changes the order of columns in the underlying database. But one might also argue the case that using an ordinal has the advantage of not breaking if someone changes the column name in the underlying database. I prefer the former argument, though, and think the readability argument for column names trumps the second argument in general. And an additional argument for names is that it is that it can “self-detect” errors. If someone does change a field name, then the code has a better chance of breaking rather than having the subtle bug of appearing to work while it reads the wrong field.

It seems obvious but maybe it is worth mentioning a usage case that has both the self-detecting error and the performance of ordinals. If you specify the SELECT list explicitly in the SQL, then using ordinals won’t be a problem because the statement in the code guarantees the order:

SELECT name, address, phone from mytable

In this case, it would be fairly safe to use ordinals to access the data. It doesn’t matter if someone moves fields around in the table. And if someone changes a name, then the SQL statement produce an error when it runs.

And one final point. I just ran a test on a provider I helped write. The test read 1 million rows and accessed the “lastname” field on each record (compared against a value). The usage of rdr[“lastname”] took 3301 milliseconds to process while rdr.GetString(1) took 2640 milliseconds (approximately a 25% speedup). In this particular provider, the lookup of the name uses a sorted lookup to translate the name to ordinal.