sql atomic exists/insert - multiple unique fields - sql

Question

Ok, so I've learned from other posts/sites, that the following style of query will execute atomically:
INSERT INTO Foo(field1, field2)
SELECT #field1, #field2
WHERE NOT EXISTS (SELECT * FROM Foo
WHERE Field1 = #field1
OR Field2 = #field2)
This would be used to satisfy a design requirement that both field1 and field2 be unique. So far, all is fine and dandy.
BUT, in the case of such a record already existing (where the query inserts 0 rows) what if I want to know WHICH field already had a record with the same value. (i.e., which of the following statements are true?
there was already an existing record where Field1 = #field1
there was already an existing record where Field2 = #field2
both of the above
You would need to know which field is at fault in order to show an error message that tells the user which field they need to change to a unique value. I could always check with a second query, but that would not be atomic. The data may have changed between the attempted insert and the 2nd query to determine which field was at fault, and the error message may not reflect the actual state of the DB.
Any ideas on how to handle this case?

Answer 1

Assumptions:
You expect most INSERTs to succeed.
You have a unique contraint such that a bad INSERT will fail.
You aren't concerned about matching NULLs.
set transaction isolation level serializable;
begin transaction;
begin try
insert into Foo ( Field1, Field2 ) values ( #Field1, #Field2 );
end try
begin catch
-- Really ought to confirm that the error was the expected one here.
select
#Field1Match = case when Field1 = #Field1 then Cast( 1 as Bit ) else Cast( 0 as Bit ) end,
#Field2Match = case when Field2 = #Field2 then Cast( 1 as Bit ) else Cast( 0 as Bit ) end
from Foo
where Field1 = #Field1 or Field2 = #Field2;
end catch;
commit transaction;

Answer 2

Based on my research (here & here), unless you have a very high confidence that over 95% of the inserts will succeed, I would use the opposite approach as HABO. Borrowing from the same assumption that you aren't concerned about matching NULLs:
DECLARE #Col1Match VARCHAR(20), #Col2Match VARCHAR(20), #msg NVARCHAR(4000);
SET ANSI_WARNINGS OFF;
BEGIN TRANSACTION;
SELECT #Col1Match = MAX(CASE Field1 WHEN #Field1 THEN 'Field1 exists:' END),
#Col2Match = MAX(CASE Field2 WHEN #Field2 THEN 'Field2 exists:' END)
FROM dbo.Foo WITH (HOLDLOCK)
WHERE Field1 = #Field1 OR Field2 = #Field2;
IF #Col1Match IS NULL AND #Col2Match IS NULL
BEGIN
INSERT dbo.Foo(Field1, Field2) SELECT #Field1, #Field2;
END
ELSE
BEGIN
SET #msg = LTRIM(COALESCE(' ' + #Col1Match + ' (' + #Field1 + ')', '')
+ COALESCE(' ' + #Col2Match + ' (' + #Field2 + ')', ''));
END
COMMIT TRANSACTION;
IF #msg IS NOT NULL
BEGIN
RAISERROR(#msg, 11, 1);
END