CppCon 2018: Ben Deane “Operator Overloading: History, Principles and Practice”

—
—
Your code would benefit from greater use of operators. Really.

Operator overloading is such an unloved part of C++. It's not new, it's not
glamorous, it's frequently maligned for "syntax abuse", and often avoided due to
performance or implementation concerns. But operators deserve more
consideration! If you're designing types and interfaces, operators have the
potential to offer concision, clarity, and flexibility, and foster intuition in
a way that functions can't.

In this talk we'll look at past, current and likely future practices, from the
humble old operator+ on arithmetic types to new uses like operator/ to
concatenate paths, and beyond. We'll cover concrete guidelines for principled
operator usage and the implementation concerns involved.
—
Ben Deane
Ben was in the game industry for 23 years, at companies like EA and Blizzard. He's always looking for useful new techniques in C++, and he | geeks out on algorithms, APIs, types and functional programming.
—

*-----*
*-----*

[arisweedler]

The fact that <=> enforces sane-ness between < <= == => > != with such ease is the best argument for it, to me. He makes it sound so simple!

[mrlithium]

I usually am able to follow talks like this, but this one was way over my head. Opened my eyes to weird maths stuff I never knew about.

[arisweedler]

Weak ordering only makes sense when you have two ways to determine equality. A filesystem can be "equal as a string" or "equal as a location". This pair of strong ordered set of operations/measurements are much easier to deal with. If only we had C++ syntax to support that. Do equality as a string on 2 filepath objects. Or do equality as a "location" on 2 filepath objects.

You get that with static cast, but this sloppy notation is what drives program designers to make mistakes because they aren't thinking about types. It's the same problem as he quoted Knuth stating, at 9:21 - not thinking about objects/operations with different mathematical rules causes errors. If it was a helpful thing to read, especially to the uninitiated, then it would be used more. (Auto actually helps with this as it eschews the need for unnecessary declarations of type - it increases the signal to noise ratio for types)

Rust, with it's "object as i32" notation surely is the same thing as "static_cast<int32_t>(object)", but one's a lot closer to what my thoughts already are. Even if C++ can't fix that situation, we can goddamn work around it with strong and weak ordering.

Syntax and making it natural expressions of thought is really important. It was a huge discovery that the 1956 compiler made apparent. Even the brightest minds will make mistakes in translating thought to code when the code doesn't map perfectly to thought. Not "well", but "perfectly". I don't think strong and weak order will every map perfectly to human thought. I think they need to be in C++ though, because it has to respect backwards compatibility.

[KvapuJanjalia]

What about getting overloaded with operators?

[trinidad]

% is the percent sign which wikipedia tells me has been in use since the 17th century, just sayin.