Where parallels cross

Interesting bits of life

Moldable Emacs: showing Clojure dependencies for the files you just modified

Too long; didn't read

I extract a Graphviz diagram for the namespaces that depend on Clojure files I just staged on Git. Also I obtain dependencies via the feature of moldable-emacs that depend on emacs-tree-sitter.

The problem

There have been a few times at work where I committed something that broke the production environment. The code I work with does not have (yet!) the reliable safety net of numerous automated tests. Add to that my rather partial knowledge of how features depend on each other and you get a production issue!

Being a startup, I do my best to spend time on developing new features. This also means I have little time to patch up the absence of tests on old features.

So I thought: how could I (at least) recognize when my change has altered some risky file?

I would love if my editor could hint that I better check things twice before opening my next PR.

And there is a solution

Well this is mostly a work in progress. My idea is that given a set of files, I can see the degree of risk in changing them. An example of set of files is those that I stage with Git and I am ready to commit. Another example that is interesting is the change-set of a PR that I am up to review for some other colleague.

The next problem was to define risk in a way I could capture with a program. After some thought I went for dependencies.

If I change a file and nothing depends on it, that is the safest I can be because I am not affecting anything. If another file depends on it, the risk increases. If two depends on it, the risk is a bit more and so on...

Say each file represents a business feature: then having 2 dependent files means I am affecting 2 features. That a file represents a feature is hardly true of course. But my intuition is that the number of dependencies somewhat reflects the probability of that being true. So if I have 10 dependent files on my change, there is a (higher) chance I have hit many features.

So let me give an example of what I got to so far:

This shows that I have changed the code of code-maat and with a bit of Elisp (that uses library code from moldable-emacs), I obtain a graph of the namespaces that depend on the files I have changed. Note that the performance_test file is ignored because my set of files is given by magit-staged-files: I left that file unstaged.

Since this looked useful but a bit flat, I refined my idea a little bit. I perceive this flat because not all dependencies are equal: some files depending on my change may have more dependencies than others. That information could help me assess risk.

So I created a complementary visualization (thanks to graph-cli) that shows the aggregate number of dependencies. Let me show you:


The smaller bars in the chart (in blue) show the number of direct dependencies for the files we have changed. The other bars (in orange) give a sense of the cumulative dependencies (i.e., the sum of files that depend on the depending files).

So with the two graphs together we know that changing the churn file in code-maat has some significant effect. In this example my modification consisted in adding a white space. The algorithm I am showing you is not smart enough to know what did you change. An improvement indeed would be to find what depends on the diff of the modified file. That is more complex though.

Also my experiment only works for Clojure files (although it would be easy to adapt for other languages).

It still an experiment because I think I could merge the two visualizations. Also, I want to make this easier to use while I am developing/reviewing PRs. I hope to do some progress and show an integrated workflow later.

And, I am still brainstorming on how to mark certain namespaces as business critical. Imagine we could easily say "these files are business critical": now if those files appear as dependencies we could show it!

So for now this is mostly a thought experiment which seems promising to me.


I showed how I am trying to visualize the risk in changing software. For now is just patchy visualization really, but I plan to refine these ideas a bit more!

Happy risking!