A New Coding Style for Switch Statements in JavaScript/TypeScript

Bryan Hughes on May 16, 2019

I've recently adopted a new coding style for switch statements in JavaScript (well, technically TypeScript), and I'm really curious to hear what yo... [Read Full]
markdown guide
 

ATM I prefer not writing switch statements and instead use conditional return statements like

const myFunction = () => {
  // ...
  return (
    (body.type === 'isBasic' && doBasicStuff()) ||
    (body.type === 'isCustom' && doCustomStuff()) ||
    new Error()
  );
};

Although this makes most sense when moving the conditions to distinct constants first, like:

const myFunction = () => {
  // ...
  const isBasic = body.type === 'isBasic';
  const isCustom = body.type === 'isCustom';
  return (
    (isBasic && doBasicStuff()) ||
    (isCustom && doCustomStuff()) ||
    new Error()
  );
};

Of course this depends on the use case. When having the necessity for more complex conditionals it might make sense to move them to distinct variables anyways and that's when this pattern shines...

 

This is a good one too. In the case of TypeScript, we do loose a bit of type checking (you can leave the default case off if the possible values are an enum IIRC), but this is good too.

I've really grown fond of using this approach in React specifically, since we can't have normal conditionals inside of JSX:

<div>
  {body.type === 'isBasic' && <BasicComponent />}
  {body.type === 'isCustom ' && <CustomComponent />}
</div>
 

I usually don't use switch statements at all, I prefer a having a dictionary object with a handler function for each case.

const handlers = {
  isBasic: ({ id }) => {
    const entry = getBasicEntry(id)
    console.log(`Processing basic entry "${entry.name}"`)
    doBasicProcessing(entry)
  },
  isCustom: ({ id }) => {
    const entry = getCustomEntry(id)
    console.log(`Processing custom entry "${entry.name}"`)
    doCustomprocessing(entry)
  },
}

const actionHandler = body => {
  const handler = handlers[body.type]
  if (handler == null) throw new Error(`Unknown type ${body.type}`)
  return handler(body)
}

Maybe it's just an idiom I brought from Python, but I think it's more natural to compose functions with this pattern than with switch statements.

 

I created a little module based on this conversation, and of course gave you a shout-out in the README :D

It's at npmjs.com/package/conditional-reduce

Code looks like this:

console.log(conditionalReduce('dog', {
  dog: () => 'Dogs are great pets',
  cat: () => 'Cat\'s are also great'
})); // Prints "Dogs are great pets"

I realized, thanks to @john_papa 's post below, that this approach also allows us to return a value in the process, so we can inline it in other expressions (or React's JSX)!

 

Wow, that went quick :D

You could also consider having the function conditionalReduce curried and pass the dictionary as the first argument, to make it easier to reuse.

const petReducer = conditionalReduce({
  dog: () => 'Dogs are great pets',
  cat: () => 'Cats are also great'
})

console.log(petReducer('dog')) // 'Dogs are great pets'
console.log(petReducer('cat')) // 'Cats are also great'

Also a great idea. I added a new helper so that folks can do either. Thanks again, this makes things cleaner!

const { curry } = require('conditional-reduce');

const dogReducer = curry({
  dog: () => 'Dogs are great pets',
  cat: () => 'Cat\'s are also great'
});

console.log(dogReducer('dog')); // Prints "Dogs are great pets"
console.log(dogReducer('bird')); // Throws 'Invalid conditional value "bird"' exception
const { reduce } = require('conditional-reduce');

console.log(reduce('dog', {
  dog: () => 'Dogs are great pets',
  cat: () => 'Cat\'s are also great'
})); // Prints "Dogs are great pets"

console.log(reduce('bird', {
  dog: () => 'Dogs are great pets',
  cat: () => 'Cat\'s are also great'
})); // Throws 'Invalid conditional value "bird"' exception
 

Oh cool, I’ve never seen this approach before! Gonna noodle on this one cause I’ve got a feeling there’s a tiny helper library in here waiting to be built to make the missing handler logic even more elegant and baked in. Thank you for sharing.

 
 

Why not let entry; at the top of the switch block?

If you're trying to set a variable from multiple place, it's not a constant.

Adding a scope, requires another stack frame. Ie increasing overhead.

 

In this example I'm not trying to set one variable from multiple places. Rather, I'm trying to set two different variables once each that never change. They actually are constants.

Each variable also has a different type in TypeScript too. There is a base class between them, but explicitly specifying the base class on the let and then casting in the function calls adds extra scaffolding that I don't think is necessary and want to avoid.

I'm not worried about adding another stack frame TBH. Once upon a time there was a performance cost to extra stack frames (and to function calls, try/catch/etc.), but V8 has been optimized enough these days that this only matters for extremely hot paths. And if this were a hot path, I'd write it pretty differently anyways, after taking the time to measure it and prove it's a hot path in need of optimization.

 

Yeah, a place I've done this is in writing switch statements in reducers. The extra scope also closes over any intermediate values you have so you don't have to coordinate variable names between blocks. I do this routinely for all switch statements all the time for this reason. The semantics are more useful and obvious with the braces.

 

I'd like to see the profiling run that shows that an extra stack frame here and there appreciably impacts performance. I promise it doesn't matter. Optimizing for readability is usually a better return on your investment than trying to optimize for performance. Also keep in mind the compiler is doing a lot of very smart optimization for you. It's better at it than we are. Of course, things that obviously matter like N+1 database queries obviously matter. Stack frames in JavaScript though? I would not worry unless you're doing something very novel.

 

Depends. Probably not significant unless it's really hot code (ex parser stap).

That likely wouldn't be a significant deal breaker, just worth considering.

My main counter is, why increase complexity, making the code both harder to read and maintain when defining entry using 'let' is much simpler/easier to understand.

To each their own. I bias heavily toward simple/maintainable code even of it's not 100% semantically perfect. Ie avoid adding complexity unless it's 100% necessary.

I think defining using a let is the harder to read and maintain option than my option, and going that path increases complexity TBH.

 

I would actually write it like this:

switch (body.type) {
  case 'isBasic':
    const basicEntry = getBasicEntry(body.id);
    console.log(`Processing basic entry "${basicEntry.name}"`);
    doBasicProcessing(basicEntry);
    break;

  case 'isCustom':
    const customEntry = getCustomEntry(body.id);
    console.log(`Processing custom entry "${customEntry.name}"`);
    doCustomprocessing(customEntry);
    break;

  default:
    throw new Error(`Unknown flag ${myFlag}`);
}

because different things warrant different variable names.

But if you cannot come up with different names, maybe these aren't actually different things and you should aim to solve the issue with a unified approach that doesn't even require a switch statement.

 

I don’t agree with this, for three reasons:

1) there are more possibilities than things are the same or are different. It’s also possible, as is the case here, that they are mostly, but not completely, the same

2) we use short-hand variables all the time that are repeated. How many callbacks have err as the first argument? Are these all the same thing? No. But we do it anyways and we consider that a best practice.

3) crafting a unified approach for this sort of case would require extra boilerplate, resulting in an over-engineered solution that obfuscates what’s actually going on.

 

1) Actually, that things are either handled the same or different is a literal dilemma without a middle ground. So in the context of their current handling they are either identical or different, regardless of their other similarities.

2) using the same shorthands twice in the same context is a recipe for confusion unless they actually mean the exactly same thing.

3) That's why I asked this as an open question. If things are actually so similar that you can handle them without too many extra cases, you should obviously do so – and if not, your variable names should reflect the difference.

I still disagree, for the reasons I stated above. I think we’ll just have to agree to disagree, but thank you for sharing your perspective!

Thanks for taking your valuable time to raise and discuss this point.

 

Reading this again (it's good, well done Bryan!).

I've been thinking about switch statements that collect a value so it can be returned at the end. imagine this ... which does not work for a few reasons.


let result = undefined;

switch (body.type) {

  case 'isBasic': {
    const entry = getBasicEntry(body.id);
    console.log(`Processing basic entry "${entry.name}"`);
    result = doBasicProcessing(entry);
    break;
  }

  case 'isCustom': {
    const entry = getCustomEntry(body.id);
    console.log(`Processing custom entry "${entry.name}"`);
    result =doCustomprocessing(entry);
    break;
  }

  default: {
    result = undefined;
  }

  return result;
}

The idea is to collect the result, then return it at the end. I'm not fond of this type of code with the let up top and then making sure each case handles a value. Any thoughts on what you do here?

 

This is a great insight! I combined this with a post above and created a little helper module: npmjs.com/package/conditional-reduce

Code looks like:

console.log(conditionalReduce('dog', {
  dog: () => 'Dogs are great pets',
  cat: () => 'Cat\'s are also great'
})); // Prints "Dogs are great pets"
 

So with that ... the code would be somehtng like this, for the example you and I used first. RIght?

let whatHappened = conditionalReduce(body.type, {
  'isBasic': () => {
    const entry = getBasicEntry(body.id);
    console.log(`Processing basic entry "${entry.name}"`);
    result doBasicProcessing(entry);
  },
  'isCustom': () => {
    const entry = getCustomEntry(body.id);
    console.log(`Processing custom entry "${entry.name}"`);
    return doCustomprocessing(entry);
  });

A few syntax issues aside, yes ;-)

It would look like:

let whatHappened = conditionalReduce(body.type, {
  'isBasic': () => {
    const entry = getBasicEntry(body.id);
    console.log(`Processing basic entry "${entry.name}"`);
    return doBasicProcessing(entry);
  },
  'isCustom': () => {
    const entry = getCustomEntry(body.id);
    console.log(`Processing custom entry "${entry.name}"`);
    return doCustomprocessing(entry);
  }
});

yeah, i figured the dev.to editor wouldn't protect me and that you'd forgive those mistakes :)

 

I prefer a variation of what you have. This is what I do in my reducers and I think it translates well in components.

I feel that this has 3 advantages:

1st, you can easily see what's going on in the switch statement and go from there.

2nd, by returning, you don't have to worry about missing a break;

3rd, essentially they are different processes and entry may have the same name, but may or may not differ in type for custom and basic...or if implementations for processing them them may change this will allow you to easily update that with separated functions.

const basicProcess = (id) => {
  const entry = getBasicEntry(id);
  return doBasicProcessing(entry);
}

const customProcess = (id) => {
  const entry = getCustomEntry(id);
  return doCustomProcessing(entry);
}

switch (body.type) {
  case 'isBasic':
    return basicProcess(body.id);
  case 'isCustom': 
    return customProcess(body.id);
  default: {
    throw new Error(`Unknown flag ${myFlag}`);
  }
 

This is definitely my second top choice, but I prefer my other approach honestly. I really don't like functions that are 2-3 lines long, unless they're dense and hard to read. IME, having a lot of small functions strung together can get hard to follow, since you're having to jump all over the file to trace the path of execution.

To be clear though, this is not a strong preference at all, it's pretty mild, and your method is definitely my second pick :)

 

All of these solutions are great, but they could benefit from having compile-time exhaustiveness checking via the never type. I describe that here: dev.to/cubiclebuddha/is-defensive-...

 

This works when each case is discreet (having its own scope), ie each case terminates the switch with a break. But one of the features of switches is that things "fall through" each case until they hit a break. I've always liked that aspect of switches, even though it's often overlooked and can make it more confusing to read/grasp.

Just curious if setting each case in a scope like your suggested style specifically prevents that fall-through or breaks (renders unexpected behavior) code between switches due to the scope not carrying over.

 

Oh man, I love when I do a little further research and find out that the thing I'm talking about has a huge debate associated with it.

en.wikipedia.org/wiki/Fallthrough_...

en.wikipedia.org/wiki/Duff%27s_device

So it may be the case that your suggested syntax adds extra reinforcement for fall-through to be avoided as an intentional design practice.

 

You beat me to my own argument 😉

That said, I myself have also used fall-through statements on rare occasions.

That that said (is that a thing?), scoping each case using {} doesn't prevent you from falling through case statements anyways.

 

Great tip! I've landed on a similar practice informed by eslint's no-case-declaration rule.

Typically I default to cases without blocks, and add them when that rule reminds me to.

 

That’s a nice approach too, as it also removes the “whoops I didn’t realize I did that” effect. Good to know, thanks!

 

As someone who comes from a long history of C++, this is how I do it in JS anyway. Forcing scope in c++ is something I've done in the middle of a code block, even, so it's pretty natural in JS to do the same.

 
 

I've actually always coded multi-line switch statements like that because of the scoping and it's easier to read IMO.

I've always been confused as to why it doesn't follow the same rules of an conditional / loop structure, requiring braces for multi-line.

 

I love this! I had this exact problem last week and I fell into the same solution. It really bothered me that I had so much duplicate code.

Glad you wrote this up. Highly readable (which I love) and a great tip for everyone.

Nicely done Bryan!

 

Is there a good reason to allocate entry at all in this scenario? My guess is you’re doing it for demonstration purposes but it’s overly verbose otherwise.

 

Correct, this was for demonstration purposes, and I agree that in practice I'd just inline the call in this case. I'll tweak the examples to make this more clear, thanks! My actual code that prompted me to write this did 3 things with the entry variable.

 
 

Simple if... else statements will not hurt if the conditional checks are short enough. It does help improve code readability. But it's your call to use which method that suits you best.

 

I personally like the extra structure that switch statements bring for iterating over possible variations in value.

code of conduct - report abuse