public interface TreeDataBinding
TreeGrid component is a visual
 representation of a tree and requires
 a Tree or ResultTree datatype passed via the TreeGrid.data attribute to
 initialize the tree view.  The Tree datatype is used
 when you want to provide all of
 the tree nodes in one shot at initialization time.  The ResultTree datatype is used
  when you want portions of the tree to be loaded on demand from the server.
  
  The simplest mechanism by which to initialize a Tree is to simply provide all the data
  up-front when the Tree itself is created.  Depending on the format of your tree data, this
 can be done by setting Tree.root or
 Tree.data.  This functionality is
 provided
  by the Tree class.
  
For examples of this type of databinding, see the following SDK examples:
  In this mode, tree nodes are loaded on-demand the first time a user expands a folder.  This
  approach is necessary for large trees.  This functionality is provided by the
 ResultTree class, which uses a DataSource to load data from the server.  Each
  DataSource Record becomes a TreeNode.
  
  When the user expands a folder whose children have not yet been loaded
  from the server (or you programmatically call openFolder() on such a node), the client
 automatically sends a DSRequest to the server to ask for all
 immediate children of
  that node.  
  
 If you have a dataset that is "parent-linked",
 that is, every node has
 a unique ID (the Tree.idField) and
 also has a property with the unique ID of it's
 parent node (the Tree.parentIdField) the tree can load child nodes by simply sending
 a DSRequest with appropriate Criteria.  Given a parent node
 with ID "225" in a tree
 where the Tree.parentIdField is
 called "parentId", the criteria would be:
  
     { parentId : 225 }
  
  The client is asking the server: "give me all nodes whose parentId is 225", which are the
  children of node 225.  
  
 If you have a DataSource that supports simple Criteria like
 the above, and your
  records have nodes with ids and parentIds, this strategy can be used by just declaring the
 tree relationship in your DataSource: the first DataSourceField.foreignKey field
  referring to another field in that same DataSource will be taken as the
 Tree.parentIdField, and the
 field to which it refers, the Tree.idField.  You
 can use the DataSourceField.primaryKey field as your Tree.idField, but it's
  only required that it hold unique values.  Note, this only works for DataSources with  a
  single primaryKey field; composite keys are not supported for this kind of tree databinding.
 See DataSourceField.primaryKey
 for more details.
  
If you have a tree where there is no convenient unique ID, for example, you have mixed types of nodes (for example, departments and employees), use one of the following approaches:
Typically two or more properties can be combined into a String that serves as a unique ID. For example, if you are loading a mixed tree of "Departments" and "Users", each of which have unique numeric IDs, you could generate synthetic node IDs like "department:353" and "user:311". Your server-side code will then receive these synthetic node IDs when the tree loads children, and you can parse the IDs, look up the appropriate object and return its child nodes.
In the case of filesystems or XML documents, you can use the full path to the file or XML element as the unique ID.
sent to the server
  If having all the properties of the parentNode would allow you to look up children, this approach may be more convenient than having to generate synthetic node IDs and parse them when looking up children.
For example, with a mixed-type tree, your server-side code may be able to quickly identify the type of the parent node be looking for specific properties, and then call methods to look up children for that type of node.
In this case there is no need to declare an idField or parentIdField.
 ResultTrees are created for you by the TreeGrid when you set
 TreeGrid.dataSource, but you
 can pass an initial dataset to a databound TreeGrid by
  setting TreeGrid.initialData.
  
 If you do not provide TreeGrid.initialData, the first DSRequest you receive will be a
  request for the nodes under root.  The id of the root node of the tree is the value of the
 rootValue attribute on the Tree.parentIdField of the Tree
 DataSource. 
  
For examples of this type of databinding, see the following SDK examples:
  When using load on demand, the Tree cannot simply check whether a node has children to
  determine whether it's a folder, and will assume all loaded nodes are folders.  To avoid
  this, you can add a boolean field to your DataSource called "isFolder" that indicates
  whether a node is a folder or not.  If you already have a boolean field that indicates
 whether a node is a folder, you can instead set Tree.isFolderProperty to the name of
 that field via TreeGrid.dataProperties.
  
  The ResultTree's DSRequests ask for the immediate children of a node only (by specifying
  parentId in the criteria). Any nodes returned whose parentId field
 value is unset or matches this criterion will be added to the tree as immediate children of the
  node. However you are also free to return multiple levels of children.  This can be done by
  simply returning a flat list of descendents with valid id's and parentId's, exactly as though 
 you were initializing a multi-level tree via Tree.data.  
  
Note that when receiving multiple levels of children, the ResultTree's assumption is that if any children are loaded for a parent, then that parent is considered fully loaded.
  When loading children for a given parent node, the ResultTree calls
 DataSource.fetchData() on its DataSource.
  For custom code that may need to reference
  the parentNode or tree in some way, the parent node whose children are being loaded is
  available on the dsRequest instance in the DataSource flow as dsRequest.parentNode, where it
 can be inspected during DataSource.transformRequest().
  
For an example of this feature, see the following SDK example:
  If some nodes in your tree have a very large number of immediate children, you can enable
 fetchMode:"paged" to load
 children in batches.  This means that
  whenever the children of a folder are loaded, the ResultTree will set
 DSRequest.startRow and endRow when requesting children from
  the DataSource.  This includes the initial fetch of top-level nodes, which are children of
  the implicit root node.
  
  As with all paged DSRequests, the server is free to ignore startRow/endRow and
  simply return all children of the node.  This allows the server to make on-the-fly
  folder-by-folder choices as to whether to use paging or just return all children.  However,
  whenever the server returns only some children, the server must provide an accurate value for
 DSResponse.totalRows.  Note that
 startRow, endRow, and 
  totalRows all refer to the array of siblings that are the direct children of a
  particular parent node (specified by the criteria or, implicitly, the root node).
  In particular, totalRows is a count of exactly how many nodes match the
  criteria; this is the same as how many total sibling children are available at that level.
  
  If the server does return a partial list of children, the ResultTree. will
 create a ResultSet to represent it rather than an array, and
 automatically request
  further children as they are accessed; typically this happens because the user is scrolling
 around in a TreeGrid which is viewing the
 ResultTree.  The value of 
 ResultTree.resultSize (which
 may be overridden via dataPageSize) will
  be passed along to any created ResultSet as well as used directly for the
  initial server request for a node's children.
  
In this mode, the server may return multiple levels of the tree as described above ("Multi-Level load on demand"), however, by default the server is not allowed to return folders that are open, as this creates a potential performance issue: consider the case of a user scrolling rapidly into an unloaded area of the tree, skipping past many nodes that have not been loaded. If the skipped nodes might have been open parents, then the only way to know what nodes should be visible at the new scroll position is to load all skipped nodes and discover how many visible children they had.
  If this performance consequence is acceptable, the restriction against returning open
  folders from the server may be lifted on a tree-wide basis by setting the
 canReturnOpenFolders property to true
  and/or on a folder-by-folder basis by setting the property named by the
 canReturnOpenSubfoldersProperty to
  true.  In this case, it is recommended to also set
 progressiveLoading to
 true to prevent
  users from causing a large number of nodes to be loaded by scrolling too far ahead in the
  tree.
  
  DsResponses from the server for a paging ResultTree should have the format:
  
{startRow:  <n> // requested start row (index within direct siblings)
  endRow:    <m> // requested end row (index within direct siblings)
  totalRows: <p> // total rows that match the criteria (all direct siblings)
  data: [
      // would have m - n entries
      // if for any folder, "isOpen" property is true, node must have the format:
      {isOpen: true,
       children: [
           // child nodes - the server decides how many to send based on paging logic
       ],
       childCount: <q> // for paging; required if more children present than returned
      }
  ]}
  startRow and endRow may differ from the values passed to
 ResultTree.getRange() (being
 farther apart) due to paging, and the value of
  endRow returned by the server may be less than that requested due to 
  totalRows not allowing enough requested rows.  However, you must provide all
  requested rows if they're available - you can't return less than the full range requested
  just because you feel it might improve performance (or other reasons).
  If any child node is open, the rules are:
childrenProperty, and
 unless they're all included, the
 childCountProperty
 must be specified
 childrenProperty must
 be specified as the empty array, or childCountProperty
  provided as 0childCountProperty may be provided even if not
 required as long as it's consistent with the childrenProperty,
  and both may also be optionally specified for closed nodes.  If the rule about providing
  children for each open node is violated, a warning will be logged and an immediate fetch
  issued for the missing children.  Note that in the above DSResponse, or in
  general when paging is active, you can't supply mixed levels of nodes in the data as is
  described in the section "Multi-Level Load on Demand" above for simpler modes of 
  ResultTree/Tree. 
  
  Your server logic must choose which nodes to return as open or closed, and intelligently
  decide how many children to provide for open nodes, but the server should try to satisfy the
  requested startRow and endRow at the top level, and the 
  architecture should ensure that the chosen page size is large enough to at least cover the
 maximum number of visible rows for the TreeGrid.  One reasonable approach might be
  to satisfy the top level node count, but also provide the children for all nested, open
  nodes, counting down, until the requested node count is also met through nested children.
  (Under such an approach, restrictions on the number of open nodes might be needed to avoid
  breaking the format rules we mentioned above when the included node count gets high.)
  
ResultTree.setCriteria()). 
 However, if ResultTree.keepParentsOnFilter is
  enabled then server filtering is required.  To illustrate with an example, consider a case
  where the ResultTree has 10,000 folders at root level and where criteria applied to their
  children would eliminate all but 20, which happen to be at the end of the 10,000.  Purely
  client-side logic would have to perform 10,000 fetch operations to check whether each
  root-level node had children before arriving at the final set of 20.
  For examples of this feature, see the following SDK example:
NOTE: trees with thousands of visible nodes are very difficult for end users to navigate. A majority of the time the best interface for showing a very large tree is to show a TreeGrid that displays just folders, adjacent to a ListGrid that shows items within those folders.
For example, the data in your email account can be thought of as an enormous tree of folders (Inbox, Sent, Drafts, Trash etc) with thousands of messages in each folder. However, none of the common email clients display email this way; all of them choose to show folders and messages separately, as this is clearly more usable.
Before starting on implementing paging within sets of children, carefully consider whether an interface like the above, or some entirely different interface, is actually a superior option. It is exceedingly rare that paging within sets of children is the best choice.