libSmalldb  v0.7
Trees in SQL

FlupdoCrudMachine has basic support for herarchical data stored in SQL database. It is intended for handling smaller trees, suitable for categories, user roles and similar stuff.

Theory

Implementation uses Nested Set approach combined with reference to parent, where reference to parent is primary information and nested sets are calculated.

See Nested set model.

Configuration

Implementation requires specifying names of columns used for indices describing tree structure.

The minimal SQL table to implement the tree is this:

CREATE TABLE `tree_node` (
  `id` int(11) NOT NULL AUTO_INCREMENT,     -- given by user
  `parent_id` int(11) DEFAULT NULL,         -- given by user; root node has NULL here
  `tree_left` int(11) DEFAULT NULL,         -- calculated
  `tree_right` int(11) DEFAULT NULL,        -- calculated
  `tree_depth` int(11) DEFAULT NULL,        -- calculated

  -- some additional columns go here ...

  -- primary key and index to quickly find a subtree
  PRIMARY KEY (`id`),
  KEY `tree_left_tree_right` (`tree_left`,`tree_right`),

  -- foreign key to parent node
  CONSTRAINT `parent_ibfk_1` FOREIGN KEY (`parent_id`)
      REFERENCES `tree_node` (`id`) ON UPDATE CASCADE
);

The default configuration is set for this structure, when different column names are used, they must be specified in state machine configuration:

{
    "class": "Smalldb\\StateMachine\\FlupdoCrudMachine",
    "table": "tree_node",
    "nested_sets": {
        "enabled": true,
        "table_columns": {
            "id": "id",
            "parent_id": "parent_page_id",
            "left": "tree_left",
            "right": "tree_right",
            "depth": "tree_depth"
        },
        "order_by": "id"
    },
    ...
}

Order of nodes in tree can be set using order_by option, which contains ORDER BY clause (SQL expression).

Implementation notes

The tree_left and tree_right columns are used to quickly select whole subtree and arrange nodes in a correct order. The tree_depth may be used while rendering the tree to correctly indent each node.

The parent_id column is the primary information, all tree_* columns are calculated from it.

The whole tree is rebuilt on every update, to ensure consistency. This approach is not suitable for very large trees, however, it is very simple and reliable.

Usage - Selects

To select a subtree, you need to know tree_left and tree_right of the parent, then you can recursively select all its children using simple select:

SELECT *
FROM tree_node
WHERE tree_left BETWEEN :parent_left AND :parent_right
ORDER BY tree_left

Or if you have only parent_id, you can use simple and fast subselect:

SELECT *
FROM tree_node
WHERE tree_left
    BETWEEN (SELECT tree_left  FROM tree_node WHERE id = :parent_id)
    AND     (SELECT tree_right FROM tree_node WHERE id = :parent_id)
ORDER BY tree_left

Usage - Updates

To recalculate whole tree (no partial updates supported) simply call FlupdoCrudMachine::recalculateTree() at the end of your transition handling method. That's all.