从MySQL的上一行中选择多个列

Question

Suppose I have a table like this:

| id |       date | name | value |
|----|------------|------|-------|
|  0 | 2017-01-14 |  foo |   one |
|  1 | 2017-01-17 |  bar |   two |
|  2 | 2017-01-18 | john |  five |
|  3 | 2017-01-19 |  doe |   ten |

(where date need not necessarily be ordered)

I want to be able to select some values of the previous row (based on date ). Such a functionality can be achieved by the following query:

SELECT 
  *, 
  (SELECT 
     name 
   FROM 
     example e2 
   WHERE 
     e2.dt < e1.dt 
   ORDER BY dt DESC 
   LIMIT 1
   ) as prev_name 
FROM example e1

with resulting table:

| id |         dt | name | value | prev_name |
|----|------------|------|-------|-----------|
|  0 | 2017-01-14 |  foo |   one |    (null) |
|  1 | 2017-01-17 |  bar |   two |       foo |
|  2 | 2017-01-18 | john |  five |       bar |
|  3 | 2017-01-19 |  doe |   ten |      john |

Now, this works just fine. However, it would be preferable if I could easily select multiple columns from the previous row, resulting in a result like:

| id |         dt | name | value | prev_name | prev_value |    prev_dt |
|----|------------|------|-------|-----------|------------|------------|
|  0 | 2017-01-14 |  foo |   one |    (null) |     (null) |     (null) |
|  1 | 2017-01-17 |  bar |   two |       foo |        one | 2017-01-14 |
|  2 | 2017-01-18 | john |  five |       bar |        two | 2017-01-17 |
|  3 | 2017-01-19 |  doe |   ten |      john |       five | 2017-01-18 |

This can of course be accomplished by simply copying the subquery (SELECT [..] FROM example e2 ...) into the query multiple times, but I guess this is not the preferable way to go. I have found several question on SO addressing either the "how to select records from a previous row" or the "how to select multiple columns using subqueries" problem, but not both. The latter problem is then mostly solved by using a JOIN statement, but I think this is not combinable with the "previous row" case. So my question is: what would be a better way to produce the last result, rather then copying a subquery for every column we need?

EDIT. As an extra constraint, that I did not include in the original question, "previous" could actually be something different from the previous row, but rather "the previous row that satisfies a condition". So suppose my table contains an extra boolean column b

| id |         dt | name | value | b |
|----|------------|------|-------|---|
|  0 | 2017-01-14 |  foo |   one | 1 |
|  1 | 2017-01-17 |  bar |   two | 0 |
|  2 | 2017-01-18 | john |  five | 1 |
|  3 | 2017-01-19 |  doe |   ten | 0 |

I would want the "previous row" to be the previous row with b = 1 , so the desired result would be:

| id |         dt | name | value | b | prev_name | prev_value |    prev_dt |
|----|------------|------|-------|---|-----------|------------|------------|
|  0 | 2017-01-14 |  foo |   one | 1 |    (null) |     (null) |     (null) |
|  1 | 2017-01-17 |  bar |   two | 0 |       foo |        one | 2017-01-14 |
|  2 | 2017-01-18 | john |  five | 1 |       foo |        one | 2017-01-14 |
|  3 | 2017-01-19 |  doe |   ten | 0 |      john |       five | 2017-01-18 |

I think this can still be accomplished by James Scott's answer, by simply only updating the variables when b = 1 , using an IF -statement, but maybe there is another solution possible in this case.

EDIT. SQLfiddle

Answer 1

Looks like a good use case for session variables if you only want the previous row, you can use ORDER BY to get different results.

SET @VDt := NULL, @VName := NULL, @VValue := NULL;

SELECT  id, @VName prev_name, @VValue prev_value, @VDt prev_dt, @VDt := dt dt, @VName := `name` `name`, @VValue := `value` `value` FROM example;

Messed this up when I first posted, note that the variables must be set after they are returned from the previous row. To reorder the columns (if desired) you can wrap this query in another that then reorders the result columns.

Let me know if you need anything else,

Regards,

James

Answer 2

Something like this will return the id of the 'previous' row.

 SELECT x.*
      , MAX(y.id) prev_id
   FROM example x
   LEFT
   JOIN example y
     ON y.id < x.id 
    AND y.b = 1
  GROUP
     BY x.id;

I'll leave the business of returning the rest of the data associated with this row as an exercise for the reader.