Const as Much as Possible (Not!)

[tad-ashlock]

While "Const as Much as Possible" is generally good practice, const should not be directly used on function declaration parameters, including member functions.

A quick example:

void foo (int x);
void foo (int const x) {return x * 2;}

void bar (char const * s);
void bar (char const * const s) {puts(s);}

Notice I said "directly". Hence s isn't const in the declaration, but is in the definition (direct), but what s points at is const in both, and must match in both (indirect).

The reason for doing this is to not leak a function's implementation details. Because whether a function treats a parameter variable as const or not is an implementation detail of the function. And if a function gets modified such that its parameter is no longer treated as const, then callers of that function shouldn't have to be recompiled.

This is a difference between C and C++. In C, the function definition's parameter type(s) must exactly match that of the function declarations. (At least it used to. I haven't actually developed anything significant with modern C.)

[tad-ashlock]

Dan Saks gives a better, and more thorough, description of this topic in the article "Top-Level cv-Qualifiers in Function Parameters" [PDF]. What I called "direct", he calls "top-level".

[IgnacioJPickering]

I assume you mean only for parameters passed "by value"? if the parameter is pass by reference then surely const is not an implementation detail, right?

[tad-ashlock]

Yes. But just to be explicit in our agreement about where const is applied: a parameter of reference type at the top level can't be const-qualified, since it's inherently const and can't be reseated. So whether that would be an implementation detail or not isn't really up for discussion. On the other hand, the target of a parameter of reference type can be const-qualified or not. But that's not the "direct" or "top-level" type, so that is not an implementation detail and therefore the declaration and definition must match there.