The comma operator is one of the lesser-known operators in C-like languages such as JavaScript and C++. Essentially, it delimits a sequence of expressions and only returns the result of the final one.
const a = 1;
const b = 2;
const c = 3;
const result = (a, b, c, 4, 5, 6, true);
console.log(result); // true
if (false, true) console.log('hello'); // hello
It's natural to ask then: when would it ever be useful to cram multiple expressions in a single line? Furthermore, even if it were useful, why would a comma-separated sequence of expressions (in a single line) be more readable and maintainable than a semicolon-separated sequence of statements (across several lines)? When should we prefer one over the other?
These are questions that I have struggled to answer over the years, but now I think I finally have an answer. In this article, I present a compelling case—perhaps the only one frankly speaking—for the comma operator.
A Motivating Example
Let's first talk about the conditional ternary operator. As seen below, if the condition
is truthy, it evaluates value
. Otherwise, it evaluates another
. There is emphasis in the key word "evaluation" here because the branches only execute when their condition is met.
const result = condition ? value : another;
For most cases, it's neat and pretty. Where it falls apart, however, is when we need to do more complex logic in between the branches before returning the conditional value. At this point, we resort to this unfortunate perversion:
let result; // Uninitialized! Yikes!
if (condition) {
// Do some complex stuff in between...
doSomething();
// ...
result = value; // Actual Assignment
} else {
// Do other complex stuff in between...
doAnotherThing();
// ...
result = another; // Actual Assignment
}
// Hopefully we didn't forget to initialize `result`!
Now there are many issues with this formulation.
- The
result
is uninitialized at first. This is not inherently evil, but an easy tried-and-tested way to avoid bugs due toundefined
is to just always initialize variables. - The initialization of
result
is literally at the bottom of the branch—far detached from its declaration. - By the end of the conditional, we better hope that
result
is surely initialized. If not us, we better hope that our teammates equally enforce that. If not now, we better hope that future developers uphold that, too!
There is a way around this limitation if we insist on using conditional ternary expressions. We just have to refactor the code into functions. That's definitely easier said than done. This gimmick gets old real quick!
function computeWrappedValue() {
// ...
return value;
}
function computeWrappedAnother() {
// ...
return another;
}
// How cumbersome!
const result = condition ? computeWrappedValue() : computeWrappedAnother();
Expression-based programming languages (such as Rust) have a more elegant solution. By reclassifying the if
statement as an if
expression, each branch can be evaluated and thus return values that can later be stored in a variable.
// A conditional ternary operator thus looks like this. Each branch
// returns a value, which is captured by the `result` variable.
// We thus ensure that `result` is always initialized by construction.
let result = if condition { value } else { another };
// If we wanted to do something more complex, we use the same syntax.
let result = if condition {
do_something();
// In Rust, the last expression without a semicolon is the value
// that will be "returned" by the overall `if` expression.
result
} else {
do_another_thing();
another
};
Can we emulate this in C-like languages? You've likely long foreseen where I'm headed with this, but yes!
A Compelling Case
What we want is a way to arbitrarily execute statements before returning a value within the ternary branches. Well, lucky for us, this is exactly what the comma operator is for.
// Parenthesized for clarity.
const result = condition
? (doSomething(), value) // evaluates to `value`
: (doAnotherThing(), another); // evaluates to `another`
The neat thing about this formulation is the fact that the branch expressions are only evaluated when necessary. We effectively emulate the behavior of expression-based programming languages. Gone are the days of ad hoc wrapper functions!
But alas, we can only go so far with this technique. You can imagine that for some sufficiently large n
, cramming n
statements into a single line already begs to be refactored into its own function. Personally, I would already reconsider by the time n > 3
. Anything higher than that is dubious construction in terms of readability.
// Maybe we should reconsider here?
const result = condition
? (x++, thing = hello(), doSomething(), value)
: (++y, thing = world(), doAnotherThing(), another);
// Okay, stop. Definitely turn back now!
const result = condition
? (
x++,
thing = hello(),
doSomething(),
doMore(y),
doEvenMore(thing),
value,
) : (
++y,
thing = world(),
doAnotherThing(),
doMore(y),
doEvenMore(thing),
another,
);
// Unless, of course, you're fine with this. It kinda does
// look like a Rust `if` expression if you squint hard enough.
Conclusion
Wrapping up, we have seen a compelling case for the comma operator: complex conditional ternary operations. The comma operator shines when the branches are short and sweet, but falls out of fashion real quick after three inlined statements. At that point, one is likely better off refactoring the code.
So should you use comma operators? Honestly... yeah! Readable code is mindful of the next reader, so as long as the comma chains are never egregiously long, I would accept—and even encourage—this coding style. If we consider the alternatives (i.e., uninitialized variables and refactored micro-functions), the comma operator is not so bad after all.
In practice, I already sprinkle my own codebases with these funny-looking comma operators. Though in fairness, I rarely have a need for multi-statement ternary conditionals anyway. But when I do, I have a cool tool in my belt that concisely expresses my intent.
To that end, I rest my compelling case for the comma operator.
Top comments (23)
I would say that, rather than a compelling case for the comma operator, your example is a compelling case for expression-based languages.
You know, you make a very strong point. 😅
My biggest concern would be that the form (a,b) exactly looks like a tuple used in almost every other languages (and even in mathematics). Using something what looks like a tuple but works like a hidden side-effect might cause confusion.
That's a totally fair argument. I will have to agree with you on this one. If there's any reason to contest the comma operator in C-like languages, this would be it.
Here is an example where I use that operator on a regular basis - reduce!
Array with objects that have a unique id reduced to a map, indexed by id:
a.reduce((acc,item) => (acc[item.id]=item,acc),{})
It’s more efficient than:
a.reduce((acc,item)=>({…acc,[item.id]:item}),{})
I do understand the appeal of one-liners, but how do you feel about this refactor?
I personally prefer it this way—albeit a bit more verbose than yours. I'd love to know your thoughts on that.
Your example is basically a map, therefore not very useful (not clean) to implement it with reduce. Just do: Object.fromEntries(a.map(x=>[x.id,x]))
However valid point, I can imagine situations where it is useful.
Oh, right! That is definitely much prettier than the
reduce
. Thanks for calling it out. 👏This feels like a corner case using a rarely used solution, and is only going to confuse other developers, especially juniors. This would likely lead to an organisation banning comma operators in their coding standards.
I could see that being the case. Though I'd like to clarify that I never advocated for its use everywhere. I was careful to note that readable code must be tasteful.
That's why my compromise is to only allow at most three inlined statements. That's what I've found to be the perfect balance between readability and bewilderment.
I think outright banning the syntax is counter-productive when the pattern can effectively and concisely communicate the intent of the code. It's like forcing crutches on us.
Instead of keeping ourselves away from these patterns, let's educate others (especially the juniors) on their proper and responsible use. Hence, I believe that banning the syntax only perpetuates the confusion around the comma operator. I find that unfortunate.
Simply speaking from experience here, I used to feel the same way about the comma operator.
Unfortunately, experience has taught me that this particular usage has almost always ended up being confusing to other developers, and unfortunately, the comma is such a small operator, that it gets easily missed in reviews and devs that are scanning code can frequently miss important things going on unless they are being extra vigilant (which is mentally taxing). This ends up being more costly in terms of readability than it may seem.
I get it, I love having code that is terse, and there are certain things that can make it this way that are much more easily identifiable, such as the double bang (!!) that forces a Boolean, and occasionally, but not quite as obvious, converting strings to numbers via the + operator (e.g. +'42'). Once understood, these patterns are much easier to interpret at a glance. Unfortunately, my experience with the comma operator has never turned out as favorable as the others. There's just something about it that isn't as recognizable at a glance.
Perhaps if you could enforce a rule that using it for this has to be done only under certain circumstances like ternaries, maybe it would be more recognizable, but most devs don't read the rules, they just look at code and figure out what they can do by following patterns they see. This is one pattern that just isn't as obvious what you should and shouldn't use it for. I think you're going to find a lot of pushback from most Sr. devs if you try to use it.
The only thing I've used the comma operator for is writing a "clever" macro to track where a variable is getting used :P
Some people at Twitter also had the same sentiment. The C macro ecosystem never fails to astound me. 😅
Haha! Very interesting :D
fun fact: (,) operator corresponds to flip const in Haskell. Unfortunately without partial application and lazy evaluation it has few true use cases. (Otherwise it has many).
In some languages, it's even possible to have some fun with the comma operator
This is so cursed. I love it!
If you do something like:
in either C or C++, many compilers will warn you that the value of the left-hand operand of
,
is discarded. To silence the warning, you need to cast it to(void)
.That aside, cramming stuff into an expression and using
,
just makes for unclear code. Just use anif
-else
instead.BTW, you missed the obvious primary use-case for
,
(at least in C or C++) which is inside afor
:where the second
,
is the comma operator and separates the++i
and++j
.Ah yes, that's true. I did omit that use case because I found it to be less compelling than the emulation of expression-based conditionals as in other programming languages. Though I will not deny that your example is pretty neat and useful, especially for
zip
-like operations.Yes, I agree, but I have raised my arguments against this: the primary one being uninitialized variables whose initialization is far detached from the declaration. If the only safe1 viable alternative for this is the comma operator (as shown in the article), then I'm all for it!
That's why for less than three inlined statements, I find the comma operator quite useful and readable for conditional ternaries. Anything more than that, though, is dubious construction that is worth refactoring into dedicated functions or otherwise. The
if
-else
construction would honestly be my absolutely last resort.By "safe", I mean code that is less likely for us now, us in the future, other developers now, and other developers in the future to forget to initialize the variable in all branches. ↩
It's a good article, in the sense that it describes how to write more cryptic code.
As a C++ developer for more than 30 years, I've always prioritized reading and understanding the code over the shortness of the expression.
How quick it is to read and understand the code is way more important than the shortness of the expression. And that's valid for more than 95% (number from the top of my head) of the production code, especially bearing in mind that the compilers are excellent in the optimization of "well written, but more descriptive than short" code.
I understand where you're coming from, which is why I limited my usage of it to at most three inlined statements.
Working within the context of conditional ternary expressions, let's consider the alternatives: (1)
if
statements with variable initialization far detached from the declaration and (2) ad hoc functions that encapsulate multiple statements.Frankly, I will always reject (1) because of how bug-prone it is as I've argued in the article. I'm willing to accept (2), but if one must write three-line ad hoc functions just to work around the comma operator, I think that's poor practice.
And so considering all other alternatives, this is how I reached my conclusion in the article. I strongly disagree that it's "cryptic code" (especially when we cap it at three inlined statements) because the alternatives are either bug-prone as in (1) or cumbersome as in (2). The comma operator can get the job done clearly and concisely in this happy medium.