Tech News Back Issues Issue: 080207

The Tree List Component: Loading Data

By David Swain
Polymath Business Systems

In the last issue of Omnis Tech News, we examined some basic properties of the Tree List component, but we only populated that component using the treedefaultlines property in the Property Manager. This is not the way to go for most real-world uses of this component. In a database application, we would most likely want to populate a Tree List component with values retrieved from the database - and do so in some meaningful way. So this month we will explore some strategies and techniques for getting database content into a Tree List field.

But first it might be helpful to see how we can get our static content from last time into a List variable as a transfer conduit...

Transfer Static Tree to List Variable

While a List variable is not the direct data source for a Tree List component, we have been provided with methods for transferring hierarchical data to and from a Tree List in a format that a List variable can handle. As a first step to learning how to populate a Tree List component with data, let's extract the contents of the static tree we set up in the last article into a List variable and examine those contents.

The Tree List component contains a method named $getnodelist() that performs this operation for us. This method requires three parameters and, on execution, appropriately defines and populates a List variable to contain an image of part or all of the contents of a Tree List field. In the order they must be given in the parameter list of the method, here is a brief description of each parameter:

Parameter definitions for $getnodelist()
listmode The first parameter determines the format of the List variable as well as the content returned by the method. It accepts either a special Omnis "Tree field" constant or the numeric equivalent of one. The two constants that we will use in this article are kRelationalList (0) and kFlatList (1). These will be explained in detail below.
noderef This is usually a notational reference to a specific node within the tree. But in the special case where we want to address the entire tree (which we do here), we supply an explicit value of 0. This parameter implies that we can extract any branch of our tree into a List variable. This becomes important as we develop techniques for populating a Tree List field when the user clicks a "closed" disclosure icon to open (or expand) a node.
listname The name of the List variable that we wish to hold our tree image. This variable must be in scope when the method is executed, but can be of any level of scope. In this article we will generally use an instance variable for our examples.

The two listmode constants mentioned above need a bit of clarification. So we can try both of them to see what they yield. If we execute the following code then after defining two appropriately named List variables of instance scope), we can examine the contents of our static Tree List from the perspective of a List variable in each of these modes:

Code to get relational and flat list content from static tree list

This method builds two lists based on the full contents of our static Tree List field: one using the kRelationalList constant and one using the kFlatList constant. We can use these views to learn the differences between these two techniques/formats. This knowledge will serve us well in making some decisions in "reversing" the process a bit later...

Relationallist vs Flatlist

As a reference, let us first review the expanded content of our Tree List field. Here we see it with all nodes expanded:

Reference image of expanded Tree List content
Fully expanded static tree list

If we count them all, there are 14 nodes in this tree. We can see the names of all the nodes in the illustration above. Now let's look at the "relational" list view of these contents:

Tree list contents using kRelationalList mode
Relational list of static tree contents

The relationallist value derived using $getnodelist() displays only the leaf nodes (those at the end of a branch that do not have children of their own), giving the complete hierarchical path to each in columns to the left of the name of the leaf node. There are only 9 leaf nodes in this tree, so that is the number of lines in the extracted List value. Parent nodes are not given a separate line in this list. Each column represents a level with the tree hierarchy. Notice that leaf nodes can be at any level in the tree. (Even a root node can be a leaf node, although this would be rare.) The name of each node at each level along that branch is listed, but we only see the names of the nodes, not any of their other property values, in this list extraction. This view summarizes the relationships among the nodes, but provides few details. Still, as we shall shortly see, this can be an effective format for quickly populating trees with nodes.

Now let's examine the "flat" list view of the Tree List contents:

Tree list contents using kFlatList mode
Flat list of static tree contents

The flatlist extraction shown here displays a line for each node, whether it contains other nodes or not. Any node that is not a root node (a node with no parent node) includes columns that detail the path back to the appropriate root node. (This is for the special case of loading the contents for the entire tree. We will see how this varies if we point the $getnodelist() method to a specific node shortly.) But the flat list view gives us some useful information beyond a more verbose listing of node paths.

This view also shows us a number of property values for each node in five additional columns beyond the one required by the deepest leaf node level. While these columns have not been given names in our List variable, the Omnis Programming manual gives us some brief details. In the order in which these columns are added, here is a little more discussion of each of those properties:

Property columns for lists derived using kFlatlist
$iconid If a node icon were being used in our Tree List, the full value of the icon id (an integer value that often includes a size constant offset) would be given in this column. Since we are not using this feature in this example, the value is 0.
$ident We assigned these values when we manually built the Tree List content for the last article. It was important to assign unique values so that we could detect precisely which node was selected when using the $currentnode() method. This property is empty if it is not specifically assigned by us - unlike most other $ident properties in Omnis Studio.
$enterable We did not make the nodes of our Tree List enterable, so this property is set to kFalse for each node. But the implication here is that it could be set to kTrue - and on a node-specific basis. If it were, the end user would be able to change the text of the label (and, therefore, the $name property value) for a node.
$expandcollapse When extracting node information from an existing Tree List, this column shows the current state of each node. (There is no real property with this name, but the name describes the state of the node represented by this column.) The value of 2 shown here is the numeric equivalent of the constant kTREEnodeExpanded. Each parent node was already expanded when we used $getnodelist() - and each leaf node is considered to be expanded as well. An unexpanded node that has at least one child would return a value of 0 for this property.
$textcolor This property gives the base text color for the label of each node. The negative integer value shown here is the numeric equivalent of the Omnis Studio system color constant kColorDefault.

Sop we see that the kFlatList technique of extracting Tree List contents into a List variable gives us a lot more detail than the kRelationalList technique. This may come in handy later - but only if we explore the use of $getnodelist() a bit further...

Examining an Individual Parent Node

So far we have used the $getnodelist() method to retrieve all the nodes from a Tree List field. But we can use that second parameter to point to a specific node. Let's see what that does for us.

We will skip the obvious intermediate steps (or we might never get past this point!) and go directly to a node that contains only leaf nodes. A good candidate is the node labeled "Sales by Customer". It contains only "Current Year" and "Previous Year" nodes (with $ident values of 121 and 122 respectively). After adding an Item reference variable that points to the tree list (to cut down on the length of our code lines), we can create a new pushbutton that retrieve our two lists for the currently selected node:

Code to retrieve relational and flat list versions of the current node

If we select the "Sales by Customer" node on our window instance and then execute this code, we will see the following two lists:

Relational and flat list images of a parent node
Relational and flat list extractions from a node containing only leaf nodes

Now we see only the child node information in our extracted list. The path to each child from its original root node is not needed because it is assumed that we already know this information (since we pointed to the parent node to retrieve the list in the first place). Note that the name of the node is now the only thing we see in the "relational" list, while we still see all the properties associated with the nodes in the "flat" list.

So What Was the Point of All This So Far?

Seeing what we can retrieve using $getnodelist() is a quick way to understanding how we can build Tree List content from our databases. You see, the $getnodelist() method has an updating counterpart: $setnodelist(). If we understand the output of the one, we are many steps closer to being able to use the features of the other!

The $setnodelist() method uses exactly the same parameter set as does $getnodelist(). We can use lists formatted exactly as we have seen to either populate an entire tree, populate a single node with children or something in between! All we have to do is to load a List variable with the appropriate information and then point $setnodelist() at the appropriate node to add that content to a Tree List field. And we can do this either for the entire tree (using a value of 0 for the node parameter) or to simply populate a given node with child nodes, just as we have so far extracted either the entire tree or just the content of a specific node.

But first we need to consider how we might retrieve that information from our database into a List variable. Remember, a tree is hierarchical in nature and we don't have any commands designed to deal with multiple hierarchical levels of data directly. There are at least two basic strategies we might use for dealing with our data hierarchically: build the entire tree content into a List variable and then transfer it to a Tree List or build each nodes contents as we need it. Each of these approaches also depends on the nature of our data in the first place - and there are at least two hierarchical structure types that we might encounter in our work. We need to understand these structures before we venture further...

Hierarchical Data Structures

There are many examples of hierarchical data structures in the world. Just because we are using a relational database doesn't mean there isn't a hierarchical aspect to our data - there usually is! We will juxtapose the two most common hierarchical approaches here:

Parent-Child Relationships

By far, the most common hierarchical aspect of a relational database is the parent-child relationship itself. While there are other relationships we might discover (or concoct) among the tables in a database, the primary-foreign key relationship is strictly hierarchical. Each parent has the potential for having multiple children, while each child has exactly one parent row in a specific parent table. Even in the special cases where we restrict a parent to only one child or allow a child without a parent, we can still consider this to be hierarchical. Here are some simple examples:

In a personnel database we might have tables for Divisions, Departments, Groups and Employees. Here the hierarchy is clear and unambiguous (as long as an employee belongs to only one group, etc.). Except for Employees, each node is assumed to be expandable. The only leaf nodes would be the Employees nodes.

In a tOmnis Softwarebase we might have tables for Courses, Classes and Students. Since there is a Many-to-Many logical relationship between Classes and Students, we would also introduce a Registrations table to mediate that relationship. Our hierarchy here would then be Courses, Classes and Registrations, but we could also retrieve the appropriate row from Students for each Registrations row added to our tree. Again, the only leaf nodes would be at the Registrations level (assuming that all Courses have at least one Class and that all Classes have at least one Student, of course).

But there is another hierarchical structure that we might come across...

Single-Table Reflexive Relationships

Occasionally we might encounter a table that itself requires hierarchical handling. This is the case when records from that table are related to records from the same table - and there is no additional table required to manage these relationships. Such a relationship is said to be a reflexive relationship. (I thought I had made up this term years ago, basing it on a linguistic principle. I was pleased to see, while researching this article, that it is truly a formal set theory term - and it is even used the way I have always described it!) Here are a couple of examples as illustrations of this type of data:

In a database that tracks Equipment, we might want to handle "subassemblies" as separate items within the same table (because we would track the same maintenance processes for a subassembly as for its parent assembly). We would include additional columns in this table for superassembly and isSuperassembly to handle these internal relationships. The superassembly column would point to the primary key value of the parent for a given row. If a row has no parent, a value of 0 would indicate that it is a "root row". We could use such root rows to populate a dynamically built tree with its initial nodes, for example. The isSuperassembly column value could be used to determine whether to give an associated tree node a disclosure icon (that is, whether the node created to represent the row would be a leaf node). In traversing the hierarchy during some iterative process, the value of this column in a particulr row also lets us know when to turn back and go down another branch of the hierarchy.

There are a number of businesses and organizations where it is important to track referrals or sponsorship lineages in a Members or Customers table. Often there are commissions or other benefits that accrue to those people who have enlisted others and we must have a way of determining their qualifications for such benefits. To do this, we must add columns for sponsorID and isSponsor. The sponsorID is given a special value for "primary" members (those who have no sponsor - or who were directly sponsored by the person who owns the business for which the Omnis program is used) and the isSponsor defaults to kFalse for each new member. When one member sponsors another, that member's isSponsor value is then set to kTrue and the row for the new person they sponsored contains their ID value in the sponsorID column. This allows us to follow a lineage up and down the chain.

An important thing to note about these single-table hierarchies is that the number of levels that may be involved is only limited by the data we collect. This means that these hierarchies can be quite deep within this single table. In building and manipulating trees based on such data, we do not have to worry about which table to access for which node level. All nodes are derived from the same table. But there is no assurance of the number of levels we may need for building our tree. So if we wish to use the kRelationalList technique for building an entire tree at once, we would need to dynamically expand the number of columns in our transfer list as we encounter deeper levels of the hierarchy. We may well decide that this is not the best path to follow...

So we now have some idea of how our data can be viewed hierarchically. Now let's examine how we might get that data into a Tree List field.

Build List for Mass Transfer to Tree List

Constructing a list to fully populate a tree list in one pass is manageable when using a multiple table hierarchy because the number of columns required for the transfer list is fixed and known ahead of time. But this still can be unwieldy in practice for a hierarchy of any complexity. Building a tree from such structures, we access a different table for each level of the tree. We can use the primary key value for a row within the table for a given level as the $ident value for the node that represents that row (assuming that we use Long integer primary keys, since that is the data type of the $ident node property...). If we know the $level of a node, we know from which table its $name and $ident values were derived and we can easily find that row again (or that rows parent or children in the appropriate table) based on these values.

Building the entire tree for a multiple table hierarchy is easy using the kRelationalList technique, especially when working with a SQL data source where we can build the relational list content with a simple JOIN on the tables involved - and including an ORDER BY clause that properly clusters the result set at each hierarchical level. Of course, we cannot provide any property values for a node other than the $name property value using this technique. But if the name of a node is unique at its level of the hierarchy, this will work just fine to identify the row for the node (since we know which table to use for that level).

Populating the full tree for a multiple table hierarchy using the kFlatList technique is a bit more complicated. We need to "walk the tree" and add a line to our transfer list for each node. While this allows us to specify property values for each node at every level of the hierarchy, it is a lengthy process using any data source. This laborious process can be performed more easily with native Omnis techniques. It is not practical for SQL.

Building an entire tree in one pass for a single table with hierarchical content can be an even more difficult proposition. The main reason that it is difficult to handle the entire hierarchy is because the number of levels - and, therefore, the number of list columns required for our transfer list - may not be known at the outset. It can also only beperformed using the "walk the tree" technique - and so is generally not considered viable for large hierarchies.

But there is a better way...

One Node at a Time

For large, complex hierarchies, building the entire tree for all the data in a hierarchy is usually not practical. In addition to the complexity of building the content for the transfer list, the final Tree List may require a tremendous amount of RAM and the user may only need to access a small portion of the tree along only one branch. It is usually much more efficient to build the tree one node at a time, starting with a set of root nodes and adding content to an individual node as the user attempts to expand that node. Using the kFlatList technique allows us to assign property values to each child node as part of the process - including a property value that forces the display of a disclosure icon for those nodes where we expect child rows may exist.

Following this strategy, we initially present our Tree List with only root nodes. So we need a constructor method to retrieve and properly format that content. The collection of such nodes will have a common characteristic () that we can use to retrieve images of the corresponding rows from our database using a simple SELECT or Build list from file process followed by $setnodelist() with a second parameter value of 0. The kFlatList technique works best for this because we can specify a $ident value for each node as a link back to the row in the database that the node represents. We can also set the expandcollapse column value in our transfer list to kExpandCollapseAlways (1), which forces a disclosure icon to appear even though there are currently no child nodes that could be exposed. This assumes that there could be child rows in the database for that node.

We then react to the user expanding and collapsing a node by building or removing the content for just that node. Adding the content for an individual node is again a simple matter. We have already placed the primary key (or some other unique identity) value in the $ident property for existing nodes, so we can know that value for the node the user wishes to expand. Armedwith that value, it is a simple matter to retrieve the appropriate information for all child rows that point to the row associated with that node.

If we can know ahead of time whether a newly introduced node will have children, we can give that node a disclosure icon. If we do not do this, the user will not be able to expand that node. Of course, the node would automatically display such an icon if there were child nodes present, but with this technique, we only want to indicate that there should be child nodes. If we cannot know this, it is safest to include the expand-collapse icon with all nodes.

Expand and Collapse Node Events

But how do we know that the user wants to expand or collapse a specific node? There are two additional Tree List events that we must use to make our node-specific tree building process work. These are the evTreeExpand and evTreeCollapse events. Theswe are events of the Tree List field itself, but each of these events is accompanied by a pNodeItem event parameter that points to the node in the tree for which the event was activated. The pNodeItem parameter is of Item reference data type, so we can use it to both retrieve the $ident value for the target node and then supply it as the second parameter of the $setnodelist() method for adding child nodes.

So using the kFlatList technique, the process for adding a new level of hierarchical content when the user expands an existing node would voil down to:

On evTreeExpand
    ; Perform tests to determine where we are in the tree and what row is represented here
    ; Perform operation that populates our transfer list with the next hierarchical level content
    Do treeref.$setnodelist(kFlatList,pNodeItem,transferList)

The method for building the next hierarchical level of content into a "flat" list is usually relatively simple, since we only need to retrieve the children for the current node and determine which of these might have children (so that we can provide a disclosure icon that the user can "expand"). Both native Omnis and SQL techniques for retrieving child row content are straightforward. For multiple table hierarchies, the level of the expanded node determines both the table from which the children will be fetched and whether a disclosure icon is required for each child node. For single table hierarchies, the value of the "end-of-branch" indicator column is used to determine whether a specific child node gets a disclosure icon.

Good Housekeeping

Of course, if we are going to add child nodes to a specific node every time it is expanded, then we must also remove all of its child nodes when it is collapsed. Otherwise we would build up multiple sets of child nodes within each node over time! Fortunately, Omnis Studio offers us the $clearallnodes() method for just this purpose.

The $clearallnodes() method is very simple. It is a method of a Tree List node rather than of the Tree List field itself, so it requires no parameters to point to the node it is to clear. (Apply directly to the tree node...) In fact, it takes no parameters at all! And it clears all levels of nodes from within the node to which it is applied. We simply apply it to pNodeItem within an evTreeCollapse event as shown here:

On evTreeCollapse
   Do pNodeItem.$clearallnodes()

This code is completely generic, so it can be pasted anywhere we need this functionality. But we do have to be careful. If we mistakenly apply this method to the Tree List field itself, it will clear all the nodes (as advertised!) from the entire tree - including any root nodes. We would have no nodes left to expand, so the Tree List on our window instance would become completely useless to us. Just a word of caution...

On the other hand, applying $clearallnodes() to the entire Tree List is most appropriate if we are updating the entire tree with new content using the kRelationalList technique at the root level. There is a use for most everything in Omnis Studio!

(Pseudo) Code Samples

I couldn't let this subject go by without giving you at least some generic sample code (or pseudo-code at least) to illustrate the techniques I advocate here. I break these down into the two basic types of hierarchical data structures already mentioned.

Here is an example of building a tree dynamically from a multiple table hierarchy. Each tables rows are represented on a different level of the Tree List field. In the conversational example given earlier in this article, the root level would be the division level, so we would initially populate the tree with just nodes retrieved from the Divisions table. Each of these is assumed to contain at least one department, so all nodes would be given disclosure icons by default.

Initial setup (most likely in the $construct method):

; Initially define transfer list to contain columns for kFlatList transfer
; for example, {nodeName,nodeIconID,nodeIdent,nodeEnterable,nodeExpandCollapse,nodeTextColor}
; retrieve root node content from database
Do $cinst.$objs.treelistfield.$setnodelist(kFlatList,0,transferList)

When a node is expanded, we need to know first which level of the tree contains that node. This determines which table will be used to retrieve child node information. This would normally be a simple selection of rows (or list build) based on the equality of the target nodes $ident value and the child tables foreign key column. Because different tables and columns are involved at the different levels, I generally use a Case block and call appropriate subroutines as follows:

$event Method for Tree List:

On evTreeExpand
    Switch pNodeItem.$level
        Case 1 ;; division
            Do method getDepartments (pNodeItem)
        Case 2 ;; department
            Do method getGroups (pNodeItem)
        Case 3 ;; group
            Do method getEmployees (pNodeItem)
        Case 4 ;; employee
            ; this level can't be expanded
    End Switch
    Redraw {structureTree}
On evTreeCollapse
    Do pNodeItem.$clearallnodes()

getDepartments Method (for example):

Calculate nodeExpandCollapse as kExpandCollapseAlways ;; or change for all lines after populating them
; Retrieve child data into the transfer list directly
; or into a separate list and copy the name and id values into the transfer list from there
Do $cinst.$objs.treelistfield.$setnodelist(kFlatList,pNodeItem,transferList)

The exact retrieval code will vary depending on the back end database and other factors. The subroutines for each level will be very similar. When retrieving the lowest level, we would set nodeExpandCollapse to 0 to suppress the disclosure icon.

For single table hierarchies, we would do things a bit differently, but not radically so. We would still initially set up the highest hierarchical level of nodes based upon appropriate criteria, then trap the evTreeExpand event to populate agiven nodewith children:

On evTreeExpand
    Do method expandGroupNode (pNodeItem.$ident)
On evTreeCollapse
    Do pNodeItem.$clearallnodes()

expandGroupNode Method:

; Retrieve child data (including name, id and end-of-branch value) into a separate list
; for each retrieved line, copy the name and id values into the transfer list
; set nodeExpandCollapse as kExpandCollapseAlways if end-of-branch warrants it (otherwise 0)
Do $cinst.$objs.treelistfield.$setnodelist(kFlatList,pNodeItem,transferList)

The main difference is that we must determine which nodes are ultimately leaf nodes so that we don't allow them to be expandable. Then again, this is only a user interface refinement that may not be a concern in some cases.

Next Time

I hope this helps you to use the Tree List component better. There is more to tell, but it will have to wait for another time. In the next article we will examine other events, features and uses of the Tree List component, including the multi-column feature of this useful tool.

 

 

© 2007 Copyright of the text and images herein remains with the respective author. No part of this newsletter may be reproduced, transmitted, stored in a retrieval system or translated into any language in any form by any means without the written permission of the author or Omnis Software.
Omnis® and Omnis Studio® are registered trademarks, and Omnis 7™ is a trademark of Omnis Software UK Ltd. Other products mentioned are trademarks or registered trademarks of their corporations. All rights reserved.

Search Omnis Developer Resources

 

Hit enter to search

X