Question

What I Have

I have an sql database with several tables interrelated by integer keys. Here are my 3 tables, the column names, and some sample data for each table. Keep in mind that I am just typing this up to give an idea, it is not a direct copy/paste from a database (and thus the formatting is meant to convey the point, not to be readable by an sql database parser.)

Table 1

ItemTable
itemID,itemName, fruitOrVeggie, Color
1, Apple, Fruit, Red
2, Orange, Fruit, Orange
3, Carrot, Vegetable, Orange

Table 2

AttributeTypesTable
attributeID,attributeName
1, Price
2, Weight
3, Diameter

Table 3

ItemAttributesTable
itemID,attributeID,attributeValue
1, 1, .75
1, 2, .5
1, 3, .7
2, 1, .9
2, 3, .7
3, 1, .3
3, 2, .5

Note how I have multiple entries for each itemID in the ItemAttributesTable – this is the part I am trying to consolidate in a new table.

What I Want

From these three tables I want to create a new table like this.

NewTable
itemID,itemName,fruitOrVeggie,Color,Price,Weight,Diameter
1, Apple, Fruit, Red, .75, .5, .7
2, Orange, Fruit, Orange, .9, , .7
3, Carrot, Vegetable, Orange, .3, .5, 

In this NewTable, itemID is a unique key so that there is only one entry per itemID – this is the goal. Note how each attributeName is now a column in this new table and how the corresponding data from ItemAttributesTable is now listed here with a single entry for each itemID (leaving a field blank if ItemAttributesTable doesn’t have an entry for that attributeID for that itemID). I do not want to have to hard code in the column names because my actual data has around a dozen columns and I want this query to be versatile enough to be able to keep using it even if an attributeName changes, I add or remove some of them, etc.

How to Get There

I’m mainly looking at the sql involved for this sort of complex query, although a shell of some sort to actually create this new table might be nice. For example, a query and then a Python script that runs that query to create the ItemAttributesTable.

The key parts are how to create a column in a new table based on an entry in another table (in this case, attributeName) and then how to properly pull the data from multiple tables to populate this new table.

Answer 1

In SQLServer2005+ you can use PIVOT operator for rotating a table-valued expression.SELECT…INTO creates a new table and inserts the resulting rows from the query into it

IF OBJECT_ID('NewTable') IS NOT NULL DROP TABLE NewTable
SELECT ItemID, ItemName, FruitOrVeggie, Color, Price, Weight, Diameter
INTO NewTable
FROM      
(      
 SELECT t.ItemID, t.ItemName, t.FruitOrVeggie, Color, attributeName, attributeValue
 FROM ItemTable t JOIN ItemAttributesTable at ON t.ItemID = at.ItemID
                  JOIN AttributeTypesTable tt ON at.attributeID = tt.attributeID
) x
PIVOT
(
 MAX(attributeValue) FOR attributeName IN ([Price], [Weight], [Diameter])
 ) p

SELECT *
FROM NewTable

Demo on SQLFiddle

OR

If you have an unknown number of columns(attributeName) to transformation, then you can use a dynamic PIVOT.

DECLARE @cols AS nvarchar(max),
        @query AS nvarchar(max)

SELECT @cols = STUFF((SELECT DISTINCT ',' + QUOTENAME(attributeName)
                      FROM AttributeTypesTable
                      FOR XML PATH(''), TYPE).value('.', 'nvarchar(max)'), 1, 1, '')

IF OBJECT_ID('NewTable') IS NOT NULL DROP TABLE NewTable                      
SET @query = 'SELECT ItemID, ItemName, FruitOrVeggie, Color, ' + @cols + 
             'INTO NewTable FROM 
             (
              SELECT t.ItemID, t.ItemName, t.FruitOrVeggie, Color, attributeName, attributeValue
              FROM ItemTable t JOIN ItemAttributesTable at ON t.ItemID = at.ItemID
                               JOIN AttributeTypesTable tt ON at.attributeID = tt.attributeID
              ) x
              PIVOT
              (
               MAX(attributeValue) FOR attributeName IN (' + @cols + ')
               ) p '

EXEC(@query)

SELECT *
FROM NewTable 

Demo on SQLFiddle