Question

I have a database setup as follows (simplified):

+----------+---------+----------+
|   path   |  val1   |   val2   |
+----------+---------+----------+
|    a/    |   two   |   cow    |
+----------+---------+----------+
|   a/b/   |   one   |   cat    |
+----------+---------+----------+
|  a/b/c   |   NULL  |   bat    |
+----------+---------+----------+

path is the primary key (in actual use, more like Smith/John/Mark/Jr).

PHP provides a path $mypath = "a/b/c";

Is there a way to build a result set of non null values, choosing from all rows whose path is part of $mypath.
That is, to search for the key that matches most exactly, find all non-null values in it, and if there are any nulls, to get them from the second-“best” key.

Eg. Mysql should check a/b/c and return a val2 = “bat”, but see that val1 is set to NULL.
It should therefore check a/b/ and set val1 = “one”;
The final result should be array('val1' => "one", 'val2' => "bat").

A friend told me there is a way with “LOCATE”, but I couldn’t work it out.
Is this doable?

And can someone who knows MySQL help me rewrite the title of this question to be more descriptive?

Thanks.

Answer 1

In conclusion:
It is probably impossible to do as requested using only SQL.
However, SQL can get an multidimensional array from the matching keys, and PHP can whittle it down to one array.

If anyone can improve this to only use SQL (as in the question posted), it would be much appreciated.

SQL:

SELECT * FROM `mytable` WHERE `path` = '$mypath' 
OR (LOCATE(`path`, '$mypath') = 1 AND RIGHT(`path`, 1) = '/') 
ORDER by length(`page`) desc;

PHP:

$results = mysql_query(... sql ...);
$pg = array();
foreach(array_keys($results[0]) as $key)
    foreach($results as $result)
        if ($result[$key]){ $pg[$key] = $result[$key]; continue 2; }

You can see this in practice at http://codepad.org/UnyEZJ6a

Notes:

• locate (used by Outis), and instr (used Nonym) are the same, except for that the order of arguments is reversed.
• right() is used in the query to ensure that the path is split only on the forward slash.
• The parenthesis used in the OR is technically redundant, as AND is evaluated before OR.
• If there is a solution in sql, I have been told it would use the group_by command.

Hope this can help someone else!