ESLint Warnings Are an Anti-Pattern

ESLint offers three settings for any given rule: error, warn, and off. The error setting will make ESLint fail if it encounters any violations of the rule, the warn setting will make ESLint report issues found but will not fail if it encounters any violations of the rule, and the off setting disables the rule.

I'd like to argue that using warn is an anti-pattern. Why? Because either you care about something, or you don't. The rule is either important to follow and violations should be fixed, or the rule is not important and developers shouldn't worry about it. Therefore, use error or off.

---

The Problem with Warnings

Now, there's no real issue with using the warn setting. The problem though is that when ESLint rule violations aren't enforced, developers tend to ignore them. This means that warnings will pile up in the ESLint output, creating a lot of noise and clutter.

So do you care about the rules that are being violated or not? If not, why do you have the rule enabled? If a rule serves no purpose and developers aren't addressing the warnings, get rid of the rule. If the rule is important, set it to error so that developers don't have the option of ignoring it.

---

One Caveat

There is one use case for the warn setting that I feel is valid, so let's address that. Remember, only a Sith deals in absolutes.

When introducing a new ESLint rule to your codebase, you may find that you have too many violations to fix all at once. In this situation, what are you to do? You want to keep your ESLint script passing, especially if you enforce it in your CI pipeline (which you should!). It may be counterproductive to add an eslint-disable-next-line comment along with a TODO comment for every violation of this rule as well.

So, if you're adding a new ESLint rule and find that for whatever reason you can't clean up the violations all at once, set the new rule to warn, at least for now. Understand though that this is a temporary setting and that it should be your goal to clean up the warnings as quickly as possible. Then, once the rule violations have been taken care of, you can change the rule setting to error.

---

Conclusion

ESLint warnings are an anti-pattern. Use only the error or off settings, and reserve using the warn setting only as a temporary stopgap.

Comments

[Riki Fridrich]

I like to use warnings in my projects for rules that are fine to break during development (e.g. no-console, no-unused-vars), but not in build. I use --max-warnings 0 option in build script. It looks something like this:

{
  "scripts": {
    "lint-dev": "eslint .src/**/*.js",
    "lint-build": "npm run lint-dev -- --max-warnings 0"
  }
}

I find this really handy. The warnings do not block me during development and debugging. But they show me where the code is incomplete and where my debug logs are. It's easy to find these places and clean them up. Then, with zero warnings tolerance, the debug code will never get into production.

Also, I use a pre-commit git hook to lint with zero warnings. This way these things never even leave my local computer.

[Zenobius Jiricek]

eslint shouldn't stop your dev hot module reload loop anyway?

[Drew Knab]

I don't agree.

A warn is simply an acknowledgement of a code style that may be dubious but should be allowed if the circumstance merits it.

There is a benefit to having a distinction between do not and ought not.

imho, eslint-disable-next-line is the noisy anti-pattern here.

[mikew]

This is confusing, because eslint-disable-next-line + a comment explaining why is exactly what you do when something "should be allowed if the circumstance merits it"

IMO, if you let warnings slide, you just end up with a lot of warnings. If no one pays attention to the warnings, they're just noise with no signal. So either fix them, or remove them.

Like others mentioned, try using --max-warnings 0 in CI builds. You can still do whatever the heck you want locally, but no warnings / errors make it into your build. It's so refreshing working on projects where there's no warnings or errors.

[Jan Küster]

Only thing where I could think of warn as usable is migrating a large codebase from one code style to another while having many contributions going on.

[Fyodor]

And it is a very important case, as this or similar kind of things happens quite a lot in the enterprise world. Business priorities cannot be easily neglected.

[Tyler Hawkins]

Love it! Someone else commented on this post suggesting to look into a tool called Betterer, and I think this is exactly what you're looking for. I had never heard of it before, but I poked around their docs this morning, and it looks like you can set rules that certain conditions should get bigger (more tests, higher coverage, etc.) or smaller (fewer ESLint rule violations, fewer TODO comments, etc.), and you can set dates for when goals must be met. Seems promising!

phenomnomnominal.github.io/bettere...

[Joel Louzado]

You might want to check out betterer which allows you to set rules on things like warnings as well and then trigger CI to fail if the number of warnings crosses some threshold.

The problem with having very strict linting is that it might slow down the team at a time where it might not be appropriate. In practice a policy of "no warnings" means that people see improvement work as being inherantly high-effort / low-reward and the refactoring piles up.

[Tyler Hawkins]

Thanks for sharing! I had never heard of Betterer before, but I spent some time this morning looking through their docs. It looks neat!

[Gulshan Aggarwal]

True, a rule should either pass or fail.

[Sunny Golovine]

In all my projects I always configure ESLint to error only, no warnings. In my mind it's either a problem or it isn't. It's either an error that needs to be fixed or it's not an error.

[Fyodor]

Remember, only a Sith deals in absolutes

This phrases should be emphasized 😅

[Dariusz Cichorski]

Just like you said, I think that 'warn' setting is useful for introducing new rules in some old/big codebase. In most cases it's good to clean up the warnings and change the rule to 'error' ASAP. 🙂

[Jakob Attkinson]

That's actually quite an awesome idea.