Inadvertent cross join occurring in SQL merge - sql

Question

I have two tables that I would like to join with the following columns:
Tables: #IndexDecomposition; Constituent.
Rows: Identifier; CUSIP; ISIN; SEDOL.
This is all part of a larger MERGE statement, and now I think that the cross join is happening on the INSERT statement:
MERGE indexdecomp.ConstituentWeighting targ
USING (#IndexConstituents src
INNER JOIN indexDecomp.Constituent c WITH (NOLOCK) on
((c.Identifier = src.Identifier) OR (c.Identifier IS NULL AND src.Identifier IS NULL))
AND ((c.CUSIP = src.CUSIP) OR (c.CUSIP IS NULL AND src.CUSIP IS NULL))
AND ((c.ISIN = src.ISIN) OR (c.ISIN IS NULL AND src.ISIN IS NULL))
AND ((c.SEDOL = src.SEDOL) OR (c.SEDOL IS NULL AND src.SEDOL IS NULL)))
ON ( targ.ConstituentId = c.Id
AND targ.AsOfDate = src.Date
AND ((targ.Weighting = src.Weighting) OR (targ.Weighting IS NULL AND src.Weighting IS NULL))
AND ((targ.TotalSharesHeld = src.TotalSharesHeld) OR (targ.TotalSharesHeld IS NULL AND src.TotalSharesHeld IS NULL))
AND ((targ.SharesOutstanding = src.SharesOutstanding) OR (targ.SharesOutstanding IS NULL AND src.SharesOutstanding IS NULL))
AND ((targ.NotionalValue = src.NotionalValue) OR (targ.NotionalValue IS NULL AND src.NotionalValue IS NULL))
AND ((targ.MarketValue = src.MarketValue) OR (targ.MarketValue IS NULL AND src.MarketValue IS NULL))
AND ((targ.MarketCap = src.MarketCap) OR (targ.MarketCap IS NULL AND src.MarketCap IS NULL))
AND ((targ.LastTrade = src.LastTrade) OR (targ.LastTrade IS NULL AND src.LastTrade IS NULL))
AND ((targ.Earnings = src.Earnings) OR (targ.Earnings IS NULL AND src.Earnings IS NULL))
AND ((targ.PeRatio = src.PeRatio) OR (targ.PeRatio IS NULL AND src.PeRatio IS NULL))
AND ((targ.Face = src.Face) OR (targ.Face IS NULL AND src.Face IS NULL)))
WHEN NOT MATCHED BY TARGET THEN
INSERT (
ConstituentID
,AsOfDate
,Weighting
,TotalSharesHeld
,SharesOutstanding
,NotionalValue
,MarketValue
,MarketCap
,LastTrade
,Earnings
,PeRatio
,Face
,ModifiedBy
,ModifiedDate
,CreatedBy
,CreatedDate
)
VALUES (
c.Id
,src.Date
,src.Weighting
,src.TotalSharesHeld
,src.SharesOutstanding
,src.NotionalValue
,src.MarketValue
,src.MarketCap
,src.LastTrade
,src.Earnings
,src.PeRatio
,src.Face
,'user'
,getdate()
,'user'
,getdate()
)
;

Answer 1

I think the only problem with your latest attempt is that it does not weed out combinations of all nulls (i.e. false positives where all four IDs are null). You can fix it by requiring that at least one of the four identifiers mus be non-null, like this:
SELECT *
FROM xTradeCapture.staging.IndexDecomposition src
INNER JOIN indexDecomp.Constituent c WITH (NOLOCK) on
(((c.Identifier = src.Identifier) OR (c.Identifier IS NULL AND src.Identifier IS NULL))
AND ((c.CUSIP = src.CUSIP) OR (c.CUSIP IS NULL AND src.CUSIP IS NULL))
AND ((c.ISIN = src.ISIN) OR (c.ISIN IS NULL AND src.ISIN IS NULL))
AND ((c.SEDOL = src.SEDOL) OR (c.SEDOL IS NULL AND src.SEDOL IS NULL))
AND (c.Identifier IS NOT NULL OR c.CUSIP IS NOT NULL OR c.ISIN IS NOT NULL OR c.SEDOL IS NOT NULL))

Answer 2

The second code snippet of yours is correct. The optimizer recognizes this pattern and converts it into an equi-join. So you don't get a cartesian product.

Answer 3

You could also use ISNULL if you have dummy values you know are not in the tables them selves. The example below assumes integers for Identifier & CUSIP and varchar for ISIN and SEDOL...
SELECT *
FROM xTradeCapture.staging.IndexDecomposition src
INNER JOIN indexDecomp.Constituent c WITH (NOLOCK) on
ISNULL(c.Identifier,-1) = ISNULL(src.Identifier,-1)
AND ISNULL(c.CUSIP,-1) = ISNULL(src.CUSIP,-1)
AND ISNULL(c.ISIN,'ABC') = ISNULL(src.ISIN,'ABC')
AND ISNULL(c.SEDOL,'ABC') = ISNULL(src.SEDOL,'ABC')