UPDATE
This problem actually turned out to be considerably more complex than I originally realized, given the requirement of repeating the entire tree for each path. I've simply deleted the old code as I don't want to add any further confusion.
I do want to keep it on record that using a recursive data structure makes this easier:
public class MyRecursiveObject
{
public MyRecursiveObject Parent { get; set; }
public string Name { get; set; }
public List<MyRecursiveObject> Children { get; set; }
}
You'll see very clearly why this is easier after reading the implementation code below:
private void PopulateTree(IEnumerable<MyObject> items)
{
var groupedItems =
from i in items
group i by i.Parent into g
select new { Name = g.Key, Children = g.Select(c => c.Child) };
var lookup = groupedItems.ToDictionary(i => i.Name, i => i.Children);
foreach (string parent in lookup.Keys)
{
if (lookup.ContainsKey(parent))
AddToTree(lookup, Enumerable.Empty<string>(), parent);
}
}
private void AddToTree(Dictionary<string, IEnumerable<string>> lookup,
IEnumerable<string> path, string name)
{
IEnumerable<string> children;
if (lookup.TryGetValue(name, out children))
{
IEnumerable<string> newPath = path.Concat(new string[] { name });
foreach (string child in children)
AddToTree(lookup, newPath, child);
}
else
{
TreeNode parentNode = null;
foreach (string item in path)
parentNode = AddTreeNode(parentNode, item);
AddTreeNode(parentNode, name);
}
}
private TreeNode AddTreeNode(TreeNode parent, string name)
{
TreeNode node = new TreeNode(name);
if (parent != null)
parent.Nodes.Add(node);
else
treeView1.Nodes.Add(node);
return node;
}
First of all, I realized that the dictionary will contain keys for intermediate nodes as well as just the root nodes, so we don't need two recursive calls in the recursive AddToTree method in order to get the "B" nodes as roots; the initial walk in the PopulateTree method already does it.
What we do need to guard against is adding leaf nodes in the initial walk; using the data structure in question, these are detectable by checking whether or not there is a key in the parent dictionary. With a recursive data structure, this would be way easier: Just check for Parent == null. But, a recursive structure is not what we have, so the code above is what we have to use.
The AddTreeNode is mostly a utility method, so we don't have to keep repeating this null-checking logic later.
The real ugliness is in the second, recursive AddToTree method. Because we are trying to create a unique copy of every single subtree, we can't simply add a tree node and then recurse with that node as the parent. "A" only has one child here, "B", but "B" has two children, "C" and "D". There needs to be two copies of "A", but there's no way to know about that when "A" is originally passed to the AddToTree method.
So what we actually have to do is not create any nodes until the final stage, and store a temporary path, for which I've chosen IEnumerable<string> because it's immutable and therefore impossible to mess up. When there are more children to add, this method simply adds to the path and recurses; when there are no more children, it walks the entire saved path and adds a node for each.
This is extremely inefficient because we are now creating a new enumerable on every invocation of AddToTree. For large numbers of nodes, it is likely to chew up a lot of memory. This works, but it would be a lot more efficient with a recursive data structure. Using the example structure at the top, you wouldn't have to save the path at all or create the dictionary; when no children are left, simply walk up the path in a while loop using the Parent reference.
Anyway, I guess that's academic because this isn't a recursive object, but I thought it was worth pointing out anyway as something to keep in mind for future designs. The code above will produce exactly the results you want, I've gone ahead and tested it on a real TreeView.
UPDATE 2 - So it turns out that the version above is pretty brutal with respect to memory/stack, most likely a result of creating all those IEnumerable<string> instances. Although it's not great design, we can remove that particular issue by changing to a mutable List<string>. The following snippet shows the differences:
private void PopulateTree(IEnumerable<MyObject> items)
{
// Snip lookup-generation code - same as before ...
List<string> path = new List<string>();
foreach (string parent in lookup.Keys)
{
if (lookup.ContainsKey(parent))
AddToTree(lookup, path, parent);
}
}
private void AddToTree(Dictionary<string, IEnumerable<string>> lookup,
IEnumerable<string> path, string name)
{
IEnumerable<string> children;
if (lookup.TryGetValue(name, out children))
{
path.Add(name);
foreach (string child in children)
AddToTree(lookup, newPath, child);
path.Remove(name);
}
// Snip "else" block - again, this part is the same as before ...
}