Create hierarchy from flat list using map/reduce - mapreduce

Question

Suppose I had a Couch instance full of documents like the following:
{"id":"1","parent":null},
{"id":"2","parent":"1"},
{"id":"3","parent":"1"},
{"id":"4","parent":"3"},
{"id":"5","parent":"null"},
{"id":"6","parent":"5"}
Is there a way using MapReduce to build a view that would return my documents in this format:
{
"id":"1",
"children": [
{"id":"2"},
{"id":"3","children":[
{"id":"4"}
]}
]
},
{
"id":"5",
"children": [ {"id":"6"} ]
}
My instinct says "no" because I imagine you'd need one pass for each level of the hierarchy, and items can be nested indefinitely deep.

Answer 1

By using only map function this can not be achieved, yes. But the reduce will have access to the whole list of documents emitted by the map functions: http://wiki.apache.org/couchdb/Introduction_to_CouchDB_views#Reduce_Functions
To implement that, you will need a robust reduce function, that should also be abled to use "rereduce" efficiently.
In the end, it might be easier to create a view, that will map each document by its parent as key. Example:
function(doc) {
emit(doc.parent, doc._id);
}
This view will allow to query the top level documents with the key "null" and with sub ids like "1", "3" or "5".
A reduce function could added to create a result like this:
null => [1, 5]
1 => [2, 3]
3 => [4]
5 => [6]
The structural tree you whished for is contained therein in a different format and can be created out there.