Photo by Nagesh Badu on Unsplash
TL;DR: ESLint should identify semantic errors which require programmer intervention to fix, while prettier should identify text formatting errors which do not require programmer intervention to fix.
The problem with Static Analysis
So there's this guy "Rice" and he said, "We literally can't know anything interesting about any program" Rice's theorem. Well dang.
That's a really big problem since we really want to know is this code any good?
Rice's theorem doesn't completely prevent static analysis, but it does tell us that literally every attempt at static analysis will be flawed in some way. This is a fundamental problem with computation, not ESLint or Prettier in particular.
So I really want everyone reading this to understand automated tools will never be able to fix your code.
But automated tools do still provide value.
The Good News
Rice's Theorem makes a distinction between syntatic and semantic properties of programs. Roughly speaking syntactic is "how it is written", whereas semantic is "how it behaves".
Most syntactic problems can be trivially identified and fixed. The most well-known of these are things like indentation and linebreaks. These things do not affect the behavior of the program, but do create a consistent syntactic standard for programmers.
It's the semantic properties that are the tricky ones.
A Clear Separation
We as fleshy non-automated humans therefore should be very happy with anything that can automate syntactic standards for us.
On the other hand, any automated tool intended to address semantic problems should be treated with suspicion. That is not to say we shouldn't use such tools but Rice's Theorem tells us that automation will never find every problem and even for the problems it does finds it will be wrong some of the time. So as far as I'm concerned human participation is required in the resolution of every semantic problem.
The Past
It used to be that ESLint was the go-to tool for both syntactic and semantic problems. And in many codebases it still is. But with the rise of Prettier, more and more of the syntactic stuff gets handled by Prettier. I strongly feel that this is a useful separation of responsibilities.
The Weakness in my position
Every system is legacy. ESLint did not design their API with Prettier in mind. But they did design their API with a very customizable plugin system in mind and included a 'fix' option that plugins could implement to provide automated corrections to problems. But as discussed above, I believe such automated fixes should only be permitted for syntactic problems. I have taken the position that syntactic problems should be the sole responsibility of Prettier to fix, but there certainly exist many ESLint plugins with 'fixable' rules which Prettier does not handle.
I'm ok with that.
The weakness of my position is that programmers might need to check more problems than strictly theoretically necessary. But I would much rather have that situation than the alternative.
A Closing Thought
If everyone swept leaves it would be easy to keep the gardens clean.
I think as tech nerds we all get very excited about being flashy and coming up with grand schemes to solve all our problems. But there is a quiet humility to faithfully carrying out even the tedious parts of our job. It is tempting to believe that the next piece of code we write will be the one that finally solves all our problems. Unfortunately, there will always be leaves falling on the grass. The sooner we accept that, the sooner we can stop being continually disappointed by automated tools. The best codebases I have seen, have been maintained by teams who don't try to show off but care deeply about every line of code. If everyone swept leaves... The job is never done, but thankfully I enjoy it 😄
The only problem is I need everyone to enjoy sweeping leaves as much as I do.
In any case, I am confident that if I persuade only some small percentage of programmers I ever meet of this way of thinking, it will have more impact than any line of code I ever write.