I am new to bitwise thing in MSSQL.

Let's suppose there's a table of favorite foods

insert int fav_foods(food_name,bitwiseVal)
values('Pasta',1)

insert int fav_foods(food_name,bitwiseVal)
values('Chicken',2)

insert int fav_foods(food_name,bitwiseVal)
values('Beef',4)

insert int fav_foods(food_name,bitwiseVal)
values('Fish',8)

insert int fav_foods(food_name,bitwiseVal)
values('Pork',16)

How do I write query to find people who selected more than one item and
selected items from "Pasta, Chicken, Beef, Pork"(but not fish)?
I hope my question is not confusing.....

Bostonasian wrote:
> I am new to bitwise thing in MSSQL.
>
> Let's suppose there's a table of favorite foods
>
> insert int fav_foods(food_name,bitwiseVal)
> values('Pasta',1)
>
> insert int fav_foods(food_name,bitwiseVal)
> values('Chicken',2)
>
> insert int fav_foods(food_name,bitwiseVal)
> values('Beef',4)
>
> insert int fav_foods(food_name,bitwiseVal)
> values('Fish',8)
>
> insert int fav_foods(food_name,bitwiseVal)
> values('Pork',16)
>
> How do I write query to find people who selected more than one item and
> selected items from "Pasta, Chicken, Beef, Pork"(but not fish)?
> I hope my question is not confusing.....
>

Your question isn't confusing but your design decision is. Why use
bitwise on something like this? If you were to use proper table design
this query would be trivial (and fast).

Zach

I tried to simply the example as much as possible, that's probably why
it didn't look that neccesary to build table like this.

I actually have survey data. Survey answer includes text, single select
multiple choice and multi-select multiple choice.

In answered data table, I currently have schema like following :

customer | question_id | answer
-----------------------------------------------
John | 1 | Pasta
John | 1 | Beef
John | 1 | Chicken
John | 1 | Pork

And I've got 2.4 million customers to manage, so I thought it'd save
some rows by using bitwise to reduce row numbers to one.

Bostonasian wrote:
> I tried to simply the example as much as possible, that's probably why
> it didn't look that neccesary to build table like this.
>
> I actually have survey data. Survey answer includes text, single select
> multiple choice and multi-select multiple choice.
>
> In answered data table, I currently have schema like following :
>
> customer | question_id | answer
> -----------------------------------------------
> John | 1 | Pasta
> John | 1 | Beef
> John | 1 | Chicken
> John | 1 | Pork
>
> And I've got 2.4 million customers to manage, so I thought it'd save
> some rows by using bitwise to reduce row numbers to one.
>

What you save in rows (i.e. disk space, which is cheap), you'll likely
lose in readability, mainainability, performance and standardization.
Search out one of Joe Celko's rants about thinking like a procedural
programmer and not a SQL/set based programmer because I think that's the
problem here.

Zach

> And I've got 2.4 million customers to manage, so I thought it'd save
> some rows by using bitwise to reduce row numbers to one.

I'll bet that disk space is much cheaper than the cost of the time you'll
spend fixing up a kludge like that :-)

Try this:

CREATE TABLE Foods (customer_id INTEGER NOT NULL REFERENCES Customers
(customer_id), food INTEGER NOT NULL REFERENCES Foods (food), PRIMARY KEY
(customer_id, food))

SELECT customer_id
FROM Foods
WHERE food IN (1,2,3,4,5) /* Pasta,Chicken,Beef,Pork,Fish */
GROUP BY customer_id
HAVING COUNT(CASE WHEN food IN (1,2,3,4) THEN 1 END) = COUNT(*)
/* Everything except fish */

--
David Portas
SQL Server MVP
--

Bostonasian (axkixx@gmail.com) writes:
> I am new to bitwise thing in MSSQL.
>
> Let's suppose there's a table of favorite foods
>
> insert int fav_foods(food_name,bitwiseVal)
> values('Pasta',1)
>
> insert int fav_foods(food_name,bitwiseVal)
> values('Chicken',2)
>
> insert int fav_foods(food_name,bitwiseVal)
> values('Beef',4)
>
> insert int fav_foods(food_name,bitwiseVal)
> values('Fish',8)
>
> insert int fav_foods(food_name,bitwiseVal)
> values('Pork',16)
>
> How do I write query to find people who selected more than one item and
> selected items from "Pasta, Chicken, Beef, Pork"(but not fish)?
> I hope my question is not confusing.....

SELECT *
FROM tbl
WHERE fav_food & (SELECT SUM(bitwiseVal)
FROM fav_foods
WHERE food_name IN ('Pasta', 'Chicken', 'Beef', 'Pork'))

But as pointed out by others, this is a poor design. You may
save disk space, but if you need to find all that selected Chicken,
you will find that you cannot have an index on bit in an integer
column, so you get awful performance.

Look at David's query, and use that instead of the above.



--
Erland Sommarskog, SQL Server MVP, esquel@sommarskog.se

Books Online for SQL Server SP3 at
http://www.microsoft.com/sql/techinf...2000/books.asp

Bostonasian wrote:
>
> I tried to simply the example as much as possible, that's probably why
> it didn't look that neccesary to build table like this.
>
> I actually have survey data. Survey answer includes text, single select
> multiple choice and multi-select multiple choice.
>
> In answered data table, I currently have schema like following :
>
> customer | question_id | answer
> -----------------------------------------------
> John | 1 | Pasta
> John | 1 | Beef
> John | 1 | Chicken
> John | 1 | Pork
>
> And I've got 2.4 million customers to manage, so I thought it'd save
> some rows by using bitwise to reduce row numbers to one.

Hi Bostonasian,

There is no need to denormalize or use bitwise operations for this. IMO,
bitwise operations are not suitable for this problem.

The database does not have to grow very fast. If you normalize all the
way through, you would get a Customers (reference) table, a Questions
(reference) table and a Answers (reference) table. All these reference
tables can have short keys, which you use in your CustomerAnswers (data)
table. If you have fewer than 64000 customers, fewer than 256 questions
and fewer than 256 (fixed) answers per question, then each row in
CustomerAnswers would be just 2+1+1 = 4 bytes (excluding the free format
text answers).

Your schema could look something like this:
CREATE TABLE Customers(CustomerID smallint PRIMARY KEY, Name
nvarchar(100))
CREATE TABLE Questions(QuestionID tinyint PRIMARY KEY, Question
nvarchar(3000))
CREATE TABLE Answers (QuestionID tinyint, AnswerID tinyint, Answer
nvarchar(200),PRIMARY KEY (QuestionID,AnswerID))

CREATE TABLE CustomerAnswers
(CustomerID smallint REFERENCES Customers
,QuestionID tinyint REFERENCES Questions
,AnswerID tinyint
,TextAnswer nvarchar(2000)
,PRIMARY KEY (CustomerID,QuestionID,AnswerID)
,FOREIGN KEY (QuestionID,AnswerID) REFERENCES Answers
)

Hope this helps,
Gert-Jan