Thomas Sojka

How to grow a tree with

This is a post about my journey to build org-parser-tree. As the name suggests, it is a parser that builds a Clojure tree structure from an org file. I would like to share some techniques that helped me creating this parser.

The Challenge

I had two projects which needed a parser for org files.

The first one is my journal, a giant org file where I jot down what I read, who I meet, what I watch and anything else that comes to my mind. Each year I calculate annual statistics, such as how many blogs I read, how many people I met and how often I met them, etc.

The second project is my personal website whose contents are defined in an org file. Each entry in this tree is a category e.g. a blog post, a talk or a course and these entries must be transformed to HTML.

In order to automate the creation of these two projects I needed the parser.

Org files

If you have never seen an org file, here is an example:

 * Headline
 ** Sub Headline
 *** Sub Sub Headline
 *** Sub Sub Headline 2

It is very similar to a Markdown file but instead of using # to create headers you use * and there are a variety of additional features for org files.

Growing trees without

I stumbled across the org-parser project. Currently it converts each line of an org file and returns a flat sequence of Clojure data structure. In the future org-parser will allow to output a format with structure information of the org file, but this feature is not implemented yet. So I decided to create a quick and dirty version for myself.

My first attempt was to create a recursive function to build up the tree. Although I was already aware about the API from this really great talk: The Art of Tree Shaping, I decided to try it first without zippers because I had never used them before and I wanted to do a quick and dirty version.

Hint: Later in this post we'll be using zippers and I am not going to explain that in detail because The Art of Tree Shaping is already a great explanation, so I really recommend you watch that talk.

I built a working recursive version, but I wasn't really happy with the code. I did some research and found the blog post Building trees with and without zippers. The author of that post had the exact same journey as me and shared his solution, building a tree with recursion and building a tree with zippers. The zipper version looked much nicer, so I decided to follow his advice and I started to use zippers as well. Here is my approach that allowed me to finish this task.

How to grow a tree

Define minimal source & target data structure

Defining clear goals for your data transformation tasks helps a lot. For me it looked like this:


  (def source "* Headline
  ** Sub Headline
  *** Sub Sub Headline
  *** Sub Sub Headline 2")


  (def target
    {:title "Headline"
     :children [{:title "Sub Headline"
                 :children [{:title "Sub Sub Headline"}
                            {:title "Sub Sub Headline2"}]}]})

During development, my source and target data structure are always visible to help me stay on track.

Create a fast feedback loop

After the plan is set, it helps to know the current progress as well. A fast feedback loop can be achieved easily in Clojure, you can send your current code to your REPL and inspect the results to get immediate feedback.

If your results are a big nested data structure, it is difficult to inspect the a textual representation. There are visual tools like REBL or Reveal that can help to visualize any Clojure data structure from your REPL in various representations. However, instead of using a general purpose tool to visualize my tree, I decided to use Rhizome because it fits my use case better. It is a tree visualization library that I set up with my target data structure as following:

  (rhizome/view-tree (comp sequential? :children) :children
                     :node->descriptor (fn [n] {:label (:title n)}))

view-tree expects a function that checks if a node has a branch, an accessor for the branch and the actual data structure to be visualized. Additionally, I provided a function for node->descriptor so that each node in my tree has a label. When I send my target data structure to the REPL, a new window opens:

A tree with four nodes and three branches

Now the feedback loop is ready, whenever I change something in the source data structure I can inspect as text or with Rhizome.

Setting up a proper feedback loop is a lot of additional work which you don't spend working on your actual problem (in this case, parsing a tree). But I think that this time is really well spent. Using fast feedback loops is my secret ingredient to stay focused, motivated and allows discovering edge cases quickly. I can't prove it, but to me it feels like after half an hour of coding on a problem, a proper feedback loop speeds up the development significantly compared to working on the problem without fast feedback.

Create the zipper

It's time to start with the actual problem: creating a zipper for the tree data structure. The API is very similar to Rhizome, the code to create a zipper looks like this:

  (zip/zipper (comp sequential? :children)
              (fn [node children] (assoc node :children children))
              {:title "root" :stars "" :children []})

Again, you need to provide a function to check for a branch and to access the branch. The next parameter is new, it provides a function to create a new node with children. The last parameter is the root node.

Now I can set up a loop using reduce to process the flat sequence of lines from org-parser of my source data structure:

     (z/zipper (comp sequential? :children)
               (fn [node children] (assoc node :children children))
               {:title "root" :level 0 :children []})
     (org-parser/org source))

process-line is called for each parsed line of the source data structure with the zipper and the actual line.

Iterate & enjoy

The only thing left is the implementation of process-line. This involves three steps:

Transform an org-parser line to match the target structure

This is a great opportunity to reuse everything we have learned so far, the source data structure from org-parser is:

[:head-line [:stars "*"] [:title "Headline"]]

and the target:

{:title "Headline", :level 1}

So setup a feedback loop and iterate, eventually you will come up with something similar to this:

  (defn transform-line [[_ & [[_ stars] [_ & title] _]]]
    {:title (str/join " " title)
     :level (count stars)})

Place the parsed line in the tree

As already mentioned, I won't explain the API in detail, since The Art of Tree Shaping is already a really great introduction.

So let's place one parsed line in the zipper. I bet you know what's happening next... source data structures:

  ;; zipper
  (z/zipper (comp sequential? :children)
            (fn [node children] (assoc node :children children))
            {:title "root" :level 0 :children []})
  ;; one parsed line
  {:title "Headline", :level 1}

target data structure:

  {:title "root",
   :level 0,
   :children [{:title "Headline", :level 1, :children []}]}

After a few iterations you should get something similar to:

  (defn place-in-tree [org-tree {:keys [level] :as headline}]
    (let [previous-level (:level (z/node org-tree))
          current-level level
          new-node (merge headline {:children []})]
        (= previous-level current-level) ;; no new level found
        (-> org-tree
            (z/insert-right new-node) ;; we add a sibling
            z/rightmost) ;; Move location to the right of the new node
        :else ;; new level found
        (-> org-tree
            (z/append-child new-node) ;; we add a child
            z/down ;; Move location down since it's a new level
            z/rightmost)))) ;; Move location to the right of the new node

Putting it all together in process-line

Now we've build all parts for the parser, next we need to combine everything. The first step is to transform each line from org-parser and then place it in the tree so process-tree looks like this:

  (defn process-line [tree org-parser-line]
    (->> org-parser-line
         (place-in-tree tree)))

and putting it together we get

  (->> (org-parser/org source)
       (drop 1) ;; org-parser outputs one item we don't need
        (z/zipper (comp sequential? :children)
                  (fn [node children] (assoc node :children children))
                  {:title "root" :level 0 :children []}))
       z/root) ;; builds the tree data structure


And that's how to grow a tree in Clojure. There are still some bugs and missing features in this minimal implementation (e.g. what happens if your org-file reduces its level), but if your feedback loops are in place you can fix this. If you don't want to grow a tree by yourself, you can use org-parser-tree. Checkout the source code to learn how to extend the tree creation with Multimethods.