watchexec make dev

..

I prefer to work in either of two ways:

1. With a REPL
2. In a watchexec make dev loop.

Using the REPL is nice for Clojure and Emacs Lisp. I’ve tried leaning on the REPL for programming Python and Haskell, but I didn’t enjoy it.

What about watchexec make dev? First, this is watchexec: https://watchexec.github.io/ Another option is find . | entr make dev, though entr doesn’t pick up files that are added on the fly. Here’s entr: https://eradman.com/entrproject/

So, what about watchexec make dev. I want to make tiny steps, and know that each step is working. That’s what I require from my feedback loop. (I define what I mean by feedback loop in Feedback loops, interface design and how stuff works.)

A good feedback loop has two characteristics:

1. The feedback cycle time is near instant.
2. The feedback given is broad.

We get narrow feedback if we’re only able to check the behavior of a narrow piece of our system. We get broad feedback when we can know that the whole system works as we want it to.

In a watchexec make dev loop, I typically either run the whole system, or run the whole test suite. So in practice, it might be watchexec make test. But those two can be run in parallel :)

Also in practice, I don’t pursue 100 % test coverage. Thus, I use the unit tests for broad system coverage. And I can use make dev to test what I’m working on specifically. Where the output is some printed lines, a file, or something that lets me perceive. Then I can refactor the “does this look okay?” logic into tests when I need to.