Clarification needed on the behaviour of count with null values

Question

Hi friends,
I am confused about the result given by the count aggregate function when dealing with null. Please can anybody clarify about that to me. Here is what i tried
CREATE TABLE Demo ( a int);
INSERT INTO Demo VALUES
(1),
(2),
(NULL),
(NULL);
SELECT COUNT(COALESCE(a,0)) FROM Demo WHERE a IS NULL; -- Returns 2
SELECT COUNT(ISNULL(a,0)) FROM Demo WHERE a IS NULL; -- Returns 2
SELECT COUNT(*) FROM Demo WHERE a IS NULL; -- Returns 2
SELECT COUNT(a) FROM Demo WHERE a IS NULL; -- Returns 0
Please can anybody explain me why the result is same for the first three select statements and why not for the last select statement. And what does COUNT(*) actually mean?
With Regards
Abhilash D K

Answer 1

There is a difference to the logic when using a column name versus "*" - which is explained in the documentation (and reading it is the first thing you should do when you do not understand a particularly query/syntax).  When you supply a column
(or expression) to the count function, only the non-null values are counted.  Isnull and coalesce will obviously replace a NULL values, therefore the result of the expression will be counted. 
1 and 2 are effectively the same - you replace a null value in your column with 0 and each query only selects rows with a null column value.  The count of non-null values of your expression is therefore 2 - the number of rows in your resultset.
3 is the number of rows in the resultset since you supplied "*" instead of a column.  The presence of nulls is irrelevant as documented.
4 is the number of non-null values in the resultset since you DID supply a column.  Your resultset had no non-null values, therefore the count is zero.