Where parallels cross

Interesting bits of life

Moldable Emacs: make everything moldable through Lisp!

Too long; didn't read

Moldable development is exciting! Here I show how I am exploring the power of this development method via Emacs.

The problem

Moldable development is a style of programming that puts meaning at the center of coding. In short as a programmer you want the meaning of your code to be available to the audiences that need it. For example, other programmers may like to see how data flow in your code, and they would typically sketch this with a diagram. In moldable development you want to derive that diagram from your code itself. This way your system shows its meaning in an accessible way for the audience. To see this in action, you want to check out anything related to GlamorousToolkit (as a starter I suggest this).

The idea that I could see code via other media than text made me curious. And when I fully realized that code is structured information, I got very excited! I thought: "What if I could extract information from code? Can I do in Emacs what GlamorousToolkit users can do?"

It is a problem indeed

Emacs is great because it can become what you need it to be. With my work on code-compass I saw how I could make Emacs a tool for code forensics. There is a real problem here though. When I coded the hotspot analysis I assumed that people understood what a hotspot diagram was communicating. I also made sure to introduce new people to the concept with blog posts (I have just been relaying the ideas of Adam Tornhill really).

The problem is that code-compass users had to learn about this view of a repository. And this view belongs to somebody else.

In moldable development the basic concept is that there is no view of software that is always useful. In other terms if you see all the software of the world as rooms in a hotel, there is no master key! You really need a special key/view to enter/understand it. This implies that the tool you use should make it easy to mold the key you have into the key you need for the next room.

Indeed, the focus of a tool like GlamorousToolkit is to make the generation of these views the default way of understanding code.

For the data flow example I mentioned above, this means that it should be easy to take your code, find the data you want to graph and link it together via the functions that transform that data. (In SmallTalk/Pharo this is easily possible because everything is an object, so once you parse your code, it is available in your editor runtime and you can query what you need.)

So you should see that my code-compass (and basically a lot of Emacs modes) tries to push on you a view that may or may not apply well to understanding your software. Maybe for your problem you need a slight variation or something totally different.

We need then the making of views obvious and frictionless. How can we make Emacs a platform for molding then?

And there is a solution

I was just lucky. While I was wondering about these things, emacs-tree-sitter popped up on my radar. I said to myself: "Isn't parsing just what I need for my proof of concept?"

After a bit of hacking and playing around, this is (an example of) what I got to!

In the video you can see how I:

  1. start from a C file from the GildedRose kata,
  2. transform that into a list of tokens that tree-sitter parsed
  3. I filter those tokens in only those that have duplicates
  4. I make a Org todo list to go through those duplicates.

When I first created this todo list view, I needed to refactor some code full of redundancy. Given the fact I could make tokens of the code, I manipulated the tree of tokens in PlayGround. After a few time I repeated the manipulation I created a view for it.

So now Emacs generates this todo list for me! If I remove some duplication, the todo list (I generate again, I am working on automatic reloading) will shorten. I could also share this list to colleagues to point out the redundancy that we need to handle.

And that is only the beginning! You saw that I used multiple views to get to that, right? Indeed, the power of moldable development lies in the fact you can compose views in always novel ways! And I already use those same views for other interesting use cases.

This is just an appetizer. I plan to make the code available, but before I want to publish another post to write down what is the design of this Emacs extension. This may take a little time, because I am moving back to the UK and the logistics require my attention XD. Among things I am also trying to learn SmallTalk/Pharo in order to bridge the Emacs and GlamorousToolkit worlds. Hopefully learning about this extension will make you curious enough to try out GT as well!


I am sure Emacs can be a solid basis to work in a moldable development style. My little experiments are already making my life much easier. If this is interesting to you, start by studying a bit the ideas behind moldable development because that will make you ready to get the most from my coming-soon extension.