Arlo’s Commit Notation (ACN) is a tool which can be wielded to solve certain problems: I use it to:

• speed up code review
• remind myself to take smaller, safe steps
• ship faster and more safely
• quickly scan commits for risk
• make rollback easier

If you are new to ACN it can be really hard to clearly see those problems and how ACN addresses them. I will try to describe it all in great detail in article format.

The “What?” of ACN

A summary of Arlo’s Commit Notation: prefix each commit message with a code indicating intention and risk.

Intentions

Risk levels

Examples:

. r Extract method Applesauce
^ r Extract method Applesauce
! r Extract method Applesauce
@ r Start extracting method with no name

About “unknown” risks

Even if you are diligent about writing tests for every behavior you add to the system, there may be behaviors you don’t know about but which users have come to depend on. You won’t have tests for these. These are “unknown” risks.

The “Why?” of the Why of ACN

This document is maybe not a great idea. When I’ve tried to explain this topic before, people are often overwhelmed and unconvinced. It is both too much detail at once while also being insufficient.

Ideally we’d be working together shoulder-to-shoulder in a pair or ensemble. I could introduce bits and pieces of these ideas incrementally as needed by the work. But if we can’t work that way (yet), maybe this document is a good enough idea to help you get started.

When you are just starting out on an ACN Learning Path you won’t be able to produce all the kinds of value that I assert can come from ACN. Don’t be too quick to dismiss ACN based on that initial experience. It will take time to understand deeply and wield effectively. For now I ask that you trust that this is real while you develop the skills. Adopt a little bit of ACN in your work, try it out, see what feels good and what feels bad, then come back here and learn a little more.

Here are some ways to use this document:

• If someone suggests your team adopt ACN and you want to know why you should consider that, this document describes some benefits that you could gain from that adoption.

• If you notice some aspect of the notation that isn’t working well for your context, you should modify the system to suit you better. This document will help make sure that modification doesn’t interfere with ACN’s ability to deliver value that you may not have experienced yet. Also, ACN surely isn’t the only way to solve these problems. My goal here is to give you a more complete list of those problems so that when you modify the system you don’t inadvertently leave some points unaddressed.

• When you first start using ACN you may not yet be able to effectively wield it for all of these problems. There are skills that take time to learn. That’s fine. Over time you can look for opportunities to grow those skills and make ACN more useful to you.

• If you want to check your progress in adopting ACN, ask if you are getting all the value described here. When you see a gap that’s an opportunity for a retrospective - what are we missing?

The Why of ACN will not make complete sense until you have had the chance to fully explore, learn, and apply all these skills. Give it time.

To Scan for Risk

The format of the lower-risk tags (. x / ^ x) makes the higher-risk tags (! X / @ X) stand out in a “oneline”-type history. Consider this example:

> git log --oneline
d8f708d @ F Send email on success
9934d95 . r extract variable
4658ae2 . r extract method.
6b19aaa ! r replace algorithm
b865312 . r introduce parameter.
ab5b785 . r make method static
203a1ad . r move method
af1a232 . r inline variable
e8aa838 . r inline variable
fe967d9 . r remove unused code

Notice that the ! and @ entries are easy to pick out.

Suppose you want to search through commit history to find the change that introduced a defect. With ACN it’s clear that you should start your search with the ! and @ entries. These higher-risk commits will be smaller than conventional commits, because the . r changes are separate and can be ignored or just skimmed lightly. This approach only as helpful as the extent to which lower risk (./^) commits are used: if most changes are ! and @ you won’t get this value from ACN.

To Expedite Review

You can get your code reviews / pull requests completed faster if you work in small increments, separate changes by category (feature, bugfix, refactoring), and indicate risks.

For example, if I split a 2-day feature into 1 day of preparatory refactoring and 1 day of behavior change, you can review my refactoring first. Obviously 1 day of changes is easier to review than 2 days of changes, just because of size. And applying your feedback on the smaller change will be less disruptive if I can apply that feedback before implementing the whole shebang.

It’s possible for refactorings to have much lower risk than features and bugfixes. By definition, refactorings should not change behavior but features and bugfixes must change behavior. Any behavior change might have unwanted business implications: if someone depends on that behavior you can break them. (If that behavior is undocumented you might have no idea that you are breaking them. See Hyrum’s Law.) . refactorings require far less scrutiny than other changes, because even unknown and untested behaviors are preserved. As a reviewer you can scale your attention to match the risk – more for @, less for r – instead of giving every line of change full scrutiny.

This mostly only matters in a solo-work environment. If your team is mob/ensemble programming when all the changes were made in full view of the team, then they don’t need the same kind of after-the-fact code review scrutiny.

To Nudge Developers

I feel a twinge of discomfort when I use @. It is a nudge to think about ways to avoid @. It appropriately carries a hint of shame. (Not too much - don’t make a toxic culture around this.)

As a developer when I am committing a change and need to pick the tag to apply, if it’s higher-risk, I will pause for a moment to think if I could have prepared the change a different way with lower risk.

• From any @ X change I might carve out a few small refactorings.

• A @ F or @ B could become ! F/! B with the help of some more tests.

• A @ r or ! r (or even ^ r) could be partially or fully replaced with one or more . r. This may require learning a new skill.

The effort to break up changes should be timeboxed. I stash my work and spend a little time carving out some smaller, lower-risk steps. At the end up my timebox, unstash what is left and call it good. The goal is better, not good.

To Reduce Business Risk

Refactorings tend to have different kinds of organizational/social risk than feature and bugfix work. If we are working on a feature or bugfix and introduce a bug, everyone will sigh and be disappointed and ask engineers to be more careful. If we introduce a bug while refactoring, the business will only see downside with no upside, and will be inclined to discourage refactoring. They’ll never say “don’t write any more features” but they might say “please stop refactoring.”

Any reduction in refactoring means technical debt grows faster, which has serious business consequences over time. You must guard against this.

The more of our refactoring that we can do towards the . r end of the risk scale, the less likely we are to trigger this bad reaction from business. As engineers we have the business responsibility of making refactoring safe so we can keep the code clean and continue to adapt to changing business needs.

To Change the Rules

You’ve probably worked on projects with a lightweight development process, maybe as a hobby or for a class or a startup. You make a change, check it in, and moments later it’s in production delivering value to customers or providing feedback to improve your next decision.

And you’ve probably worked on projects that, in response to the pain of production defects, have created layer after layer of heavyweight, bureaucratic paperwork every time you ship to prod.

In an organization that understands ACN, and with a team that is able to wield ACN skillfully, it makes sense to change those rules. For example, if a Director’s sign-off is currently required for every deployment to production, you instead might say that lower-case-only changes executed by a pair or ensemble can be shipped with only a Senior engineer’s sign-off. You might consider doing the same for ^-only changes (no ! or @).

It may take time to build trust and shift these rules. Make a small change to the rules, live with it for a little while, see how it plays out, then make another small change.

With a special case rule in place, developers who work in these lower-risk ways will get an “unfair advantage” over developers working conventionally. That’s great - it aligns incentive/reward with benefit to the business. Managers should make a point of praising increased productivity and reduced risk outcomes when they see it, to help shift cultural norms and expectations.

(If you are an engineer, ask your manager to do this in front of the team. Give them your best pitch. If they decline, IMO it’s time to switch managers.)

Changing the rules to smooth the path for lower-risk changes will increase productivity, improve latency, and reduce business risk. It’s win/win/win.

To Roll Back Easily

In addition to separating changes by risk and intention, we can also order them to put lower-risk items (especially refactorings) first.

Suppose you have broken up a large feature change in to a sequence like this:

8: @ F Send email on success
7: . r extract method
6: ! r replace algorithm
5: . r introduce parameter.
4: . r make method static
3: . r move method
2: . r inline variable
1: . r inline variable

We might ship changes 1 through 5 to production first. Because they contain only lower-case changes, it’s extremely unlikely they will break anything.

Then we might ship changes 6 through 8 together. Or perhaps we ship 6, 7, and 8 separately, to further compartmentalize risk - it depends on how hard it is to ship in your context.

If something blows up, we can roll back to 5 and then work forward with more care and safety.

If a ! or @ change does blow up in prod, this is a stronger nudge to look for ways to move further towards lower-case-only changes.

Hint: “refactoring to clean up the mess that is already there” should come before “refactoring in preparation for the new feature or bugfix”, since the former are more likely to be worth keeping even if plans change.