In languages with static and convenient type systems, I try to instead encode units as types. With clever C++ templating, you can even get implicit conversions (e.g. second -> hour) and compound types (e.g. meter and second types also generate m/s, m/s^2 and so on).

I think some of the modern languages handle this pretty well. Rust has algebraic data types thanks to its brilliant use of enums. Go has a similar type system. Taking the elapsedTime example from the post, for solving this duration related problem, a Rust programmer would use Duration::from_millis(millis) or Duration::from_secs(secs) and forget about the unit. It’s a duration, that’s what you wanna care about.

Those are just types. You shouldn’t write types in the names. It’s called Hungarian Notation, but it’s just redundant. If you need to check the type of a variable, hover over it and your IDE should tell you that temperatureThreshold is type DegreesCelsius. No need to add extra cruft. There’s also a question of how specific everything needs to be.

It’s also especially problematic if you later refactor things. If you change units, then you have to rename every variable.

Plus, variables shouldn’t really be tied to a specific unit. If you need to display in Fahrenheit, you ideally just pass temperatureThreshold and it converts types as needed. A Temperature type that that has degreesF() and degreesC() functions is even cleaner. Units should just be private to the type’s struct.

That seems akin to commenting. The problem with this approach is that text is not code. It’s very easy to forget to change text. In that case it becomes the worst of both worlds, you have a variable name that actually misleads you.

Much safer than this is to encode this kind of information into the code itself in such a way the program won’t compile of the types are incorrect.