Stop Using So Many Damn Ternaries

A ternary operator, for those of you who don’t know, is an operator with three parts—hence ternary operator instead of binary operator. They look like this:

  cond ? a : b

If cond is true, then the expression evaluates to a, otherwise it evaluates to b. They’re very useful in some situations, such as when you want to conditionally assign to a variable.

Contrast this

let a;
if (cond) {
  a = foo();
} else {
  b = bar();
}

With this

const a = cond ? foo() : bar();

Note that we can now use const, as a is never reassigned. Always a nice benefit.

Despite these benefits, ternaries have become rather overused, especially in the React world. This is partially due the fact that JSX only allows expressions, which precludes JavaScript’s traditional imperative if statements from being used. Multitudes of developers, seeing that if statements don’t work inside JSX, simply substitute a ternary operator for the if statement. This is not good.

First there’s the case where you can use a better tool. For instance, if you need to render something conditionally, say you have Sidebar and an isSidebarVisible variable.

Some people will write the following

{isSidebarVisible ? <Sidebar /> : null}

While this is nice and declarative, it’s also redundant. This works just fine.

{isSidebarVisible && <Sidebar />}

Second there’s the cases where one can substitute an if statement quite easily.

<div className={classes.App}>
{isUserLoggedIn
 ? isAdmin
 ? <AdminPage />
 : <UserPage />
 : <LoginPage />}
</div>

Becomes

if (isUserLoggedIn) {
  if (isAdmin) {
   return (
     <div className={classes.App}>
       <AdminPage />
     </div>
   );
  }
  return (
    <div className={classes.App}>
      <UserPage />
    </div>
  );
}
return (
  <div className={classes.App}>
	<LoginPage />
  </div>W
); 

Yes, yes, the wrapper div is being repeated. If it truly behooves you, the inner component can be assigned to a variable. A lot of people forget (or don’t know) that JSX is just syntactic sugar for function calls.

let innerComponent;
if (isUserLoggedIn) {
  if (isAdmin) {
    innerComponent = <AdminPage />
  } else {
    innerComponent = <UserPage />
} else {
  innerComponent = <LoginPage />
}

return <div className={classes.App}> {innerComponent} </div>

One may ask why ternaries are so terrible. After all, the “better” code is a lot more verbose, while the ternary code has a nice compactness to it.

Ternaries aren’t necessarily bad, but they can easily grow out of hand.

{ isUserLoggedIn 
  ? isAdmin
  ? <AdminPage />
  : isUserVerified
  ? <UserPage />
  : <VerificationPage />
  : <LoginPage /> }

Fun question, try evaluating this for true, false, true. This is hard to read for a few reasons. It’s hard to figure out the evaluation flow. It’s hard to figure out which condition goes with what branch. Finally and most importantly, the states are implicit.

Control flow is inherently about dealing with different states in your code. If something is in state A, execute this code. If something is in state B, execute this code. If statements, returns, etc. are all tools to map states to code. What if we made this state explicit?

In this example, we’re conditioning on user state. In ahem more civilized languages, we could simply make an enum/discriminated union type and pattern match on it.

enum UserState {
  LoggedOut,
  Admin,
  Unverified,
  Verified
}

match user_state {
  UserState::LoggedOut => LoginPage,
  UserState::Admin => AdminPage,
  UserState::Verified => UserPage,
  UserState::Unverified => VerificationPage
}

But if we’re using a language that doesn’t have pattern matching, we have to be a little smarter. For instance, we could use string literals in JavaScript with an object.

const USER_STATES = {
  LOGGED_OUT: "USER_STATES_LOGGED_OUT",
  ADMIN: "USER_STATES_ADMIN",
  VERIFIED: "USER_STATES_VERIFIED", 
  UNVERIFIED: "USER_STATES_UNVERIFIED"
};

const userStates = {
 [USER_STATES.LOGGED_OUT]: <LoginPage />;
 [USER_STATES.ADMIN]: <AdminPage />;
 [USER_STATES.VERIFIED]: <UserPage />;
 [USER_STATES.UNVERIFIED]: <VerificationPage />;
}

<div className={classes.App}> { userStates[myUserState] } </div>

Granted, we need to figure out how to compute the userState. And for that, you’ll need to use some if statements. But once we calculate this state, we can store it in a central location, either context or Redux, and use the same pattern of an object literal whenever we need to condition on userState.

const secretPages = {
 [USER_STATES.LOGGED_OUT]: <LoginPage />;
 [USER_STATES.ADMIN]: <SuperSecretPage />;
 [USER_STATES.VERIFIED]: <KindaSecretPage />;
 [USER_STATES.UNVERIFIED]: <NotSecretPage />;
}

<div className={classes.SecretPage}> {secretPages[myUserState]} </div>

What’s nice about this approach is that it puts the states up front. When we condition on userState, we’re now encouraged to handle all the cases (in languages with actual enums/pattern matching this would be forced at compile time). And by emphasizing states, we’ve exposed the underlying data structure: a state machine. Which brings Rob Pike’s quote on data structures vs algorithms to mind.

Data dominates. If you’ve chosen the right data structures and organized things well, the algorithms will almost always be self-evident. Data structures, not algorithms, are central to programming.

Given a complicated algorithm with a lot of boolean logic such as if (a || b && c), consider refactoring it into a series of reasonings about state. Then you won’t need to trace through the program flow: it’ll be obvious.