# An obvious guideline on function-call style

Warning: This blog post is glaringly obvious!

The other day, in yet another discussion of how to get keyword parameters into C++, I pointed to Titus Winters’ excellent guideline about how to craft an overload set:

Use overloaded functions only if a reader looking at a call site can get a good idea of what is happening without having to first figure out exactly which overload is being called.

This guideline comes verbatim from Titus’s talk “Modern C++ Design” (CppCon 2018), which took it verbatim from Google’s C++ Style Guide, which I suspect may have taken it from Titus. :)

The Google style guide refers back to this rule in several places, such as in the sections on default function arguments and operator overloading.

It struck me that this guideline could probably be generalized as follows:

Write your function calls so that the reader knows roughly what’s going to be done with each argument and what’s going to happen to its value.

This generalization encompasses a lot of other little style rules:

Name functions with verb phrases.

x.setLevel(y) is better than x.level(y). (What’s going to be done to x?) repair(c) is better than garage(c). (What’s going to be done to c?)

Name functions after their purpose, and clearly.

s.trim_in_place() is better than s.clean(). (What’s going to be done to s? Is this a mutating method, or a query?)

Never pass by non-const T&. Pass out-parameters by pointer.

getline(cin, &s) is better than getline(cin, s). trim(&s) is better than trim(s). (Will s keep its old value?) Our default assumption is that parameters that look like they’re being passed by value are being passed by value (or, as an optimization, by const&). If you write y = trim(x), and that causes x’s value to be modified, your reader will be surprised.

Use overloading only for functions that all do roughly “the same thing.”

The STL notably violates this rule when it comes to the constructor overload set: it’s not obvious what std::string("abc", 2) and std::string(2, 'a') really have in common, other than “creating a new string object” — which could be said of any kind of function with return type string. The syntax string::n_copies_of(2, 'a') would arguably have been clearer than string(2, 'a'). See “Is your constructor an object-factory or a type-conversion?” (2018-06-21).

Avoid functions that take lots of parameters, especially lots of parameters of the same type one after another.

I refer to this specific failure mode as the “boolean parameter tarpit” in “Default function arguments are the devil” (2020-04-18).

Instead of x.print(msg, /*trailingNewline=*/false, /*wrapLines=*/true), for almost all real-world use-cases, it ends up being a good idea to create completely distinct names for the entry points you intend to support: x.print(msg), x.printWithNewline(msg), x.printWrapped(msg), x.printWrappedWithNewline(msg).

In my experience, the boolean parameter tarpit is the main thing that makes people think they want named parameters in C++. People look at x.print(msg, /*trailingNewline=*/false, /*wrapLines=*/true) and immediately want to rewrite it as something like x.print(msg, trailingNewline: false, wrapLines: true). Indeed, that would help us adhere to the rule that function calls should clearly indicate what’s going to happen with each argument. But as long as C++ doesn’t support that syntax, I’d recommend using the solution above, or even a simple integral bitmask like x.print(msg, PO_TrailingNewline | PO_WrapLines).

So: I claim that this high-level guideline encompasses a whole lot of little style rules that apply to function call syntax in C++ (and, in fact, in any language).

Write your function calls so that the reader knows roughly what’s going to be done with each argument and what’s going to happen to its value.

Now, the enterprising reader will notice that really this is still a specific case, and that the above guideline can be generalized as follows:

Write your code so that the reader knows roughly what’s going on.

Kernighan and Plauger’s The Elements of Programming Style — an excellent, excellent book, recommended reading for everyone at least once in their career — starts out with two extremely high-level versions of this rule:

Write clearly — don’t be too clever.

and

Say what you mean, simply and directly.