← Back to context

Comment by gorgoiler

4 hours ago

All software anneals into a final form that is incontrovertibly correct. Each of these OpenBSD rewrites feels like the realization of such a final form!

Code is like a math proof: sketched first on a napkin, then on a blackboard, and then finally typeset in a paper. Each step tidies up the ideas to improve the communication of intent, and the final version should be self evidently stable and/or correct.

In the old days you’d see code stability emerge as the “v2” edition of some piece of software. Mozilla to Phoenix to Firebird/fox. Linux 2.2 to 2.4. Python 2.6 to 3.x. The design patterns are carried over but the implementations are revamped for more stable, more legible, and more maintainable code.

I don’t mind that vibe coding is the latest form of this phenomenon. We have all been “vibe coding” for decades really. Code like this crap:

  T = “hello world”

  def foo_2():
    FONT = “Perpetua.ttf”
    w = text(T, Font)
    w2 = w.translate((50,0,0))
    w3 = w.translate((0,0,20))
    show(w3)

…gets eventually rewritten, once it works, into:

  def render(phrase: str, font: Font) -> Shape:
    return w.text(phrase, font)

  font = Fonts.load(“Perpetua”)
  text = render(“hello world”, font).translate((50, 0, 20))
  show(text)

Before, we’d hack a v0, tidy it up sufficiently for it to be worth of review as v1, then, come back much later and rearrange the innards (in a far more sensible way) as v2.

With LLMs — especially in the hands of those who can’t read or won’t read the actual code — we are seeing a lot more version zeroes in the world. Thank you OpenBSD for giving us, albeit surprisingly for rsync, a nice v2.