I have a tree structure in a table and it uses materialized paths to allow me to find children quickly. However, I also need to sort the results depth-first, as one would expect with threaded forum replies.
id | parent_id | matpath | created
----+-----------+---------+----------------------------
2 | 1 | 1 | 2010-05-08 15:18:37.987544
3 | 1 | 1 | 2010-05-08 17:38:14.125377
4 | 1 | 1 | 2010-05-08 17:38:57.26743
5 | 1 | 1 | 2010-05-08 17:43:28.211708
7 | 1 | 1 | 2010-05-08 18:18:11.849735
6 | 2 | 1.2 | 2010-05-08 17:50:43.288759
9 | 5 | 1.5 | 2010-05-09 14:02:43.818646
8 | 6 | 1.2.6 | 2010-05-09 14:01:17.632695
So the final results should actually be sorted like this:
id | parent_id | matpath | created
----+-----------+---------+----------------------------
2 | 1 | 1 | 2010-05-08 15:18:37.987544
6 | 2 | 1.2 | 2010-05-08 17:50:43.288759
8 | 6 | 1.2.6 | 2010-05-09 14:01:17.632695
3 | 1 | 1 | 2010-05-08 17:38:14.125377
4 | 1 | 1 | 2010-05-08 17:38:57.26743
5 | 1 | 1 | 2010-05-08 17:43:28.211708
9 | 5 | 1.5 | 2010-05-09 14:02:43.818646
7 | 1 | 1 | 2010-05-08 18:18:11.849735
How would I work that out? Can I do that in straight SQL (this is PostgreSQL 8.4) or should additional information be added to this table?
Update: trying to explain sort criteria better.
Imagine that id ‘1’ is the root post to a forum and everything with a ‘matpath’ beginning with ‘1’ is a child of that post. So ids 2 through 5 are direct replies to 1 and get matpaths of ‘1’. However, id 6 is a reply 2, not directly to 1, so it gets a matpath of 1.2. This means that for a threaded forum with proper nesting, with all ids shown in the tables, the structure of the forum would look like this, hence the ordering requirement:
* id 1 (root post)
* id 2
* id 6
* id 8
* id 3
* id 4
* id 5
* id 9
* id 7
I typically create an additional columnn for this, called something like
SortPath. It would contain the data that you need to sort by, concatenated together. That column would be of typevarchar, and get sorted as a string. Something like this:A couple of things to note here:
sortpathwill be sorted as a string, so it is important all dates have the same length for it to correctly sort. E.g., observe how2010-05-08 17:38:57.26743has an extra zero added in thesortpathcolumn.sortpath, but it is not inmatpath. I would prefer to see it in both.sortcolumnas well. This is so that if you ever want to query for more than one forum at a time (you probably won’t), then it will still sort correctly.