Programming

I absolutely agree. But:

• sometimes you need to modify existing code and you can't add the types necessary without a giant refactoring
• you can't express units with types in:
  • JSON/YAML object keys
  • XML tag or attribute names
  • environment variable names
  • CLI switch names
  • database column names
  • HTTP query parameters
  • programming languages without a strong type system

Obviously as a Hungarian I have a soft spot for Hungarian notation :) But in these cases I think it's warranted.

There are plenty of times where the type is just something generic like an integer and making a wrapper type is not worth the effort and this is a useful approach.

IIRC F# even has built-in support for units.

A good example is Go's time package. You'd normally express durations like 5 * time.Second and the result is a time.Duration. Under the hood, it's just an int64 nanoseconds, but you'd never use it as a plain nanoseconds. You'd instead use it like d.Seconds() to get whichever unit you desire.

I like how in Nim you can create a type and then overload default operators to support custom types (operators are just functions with 2 arguments in Nim), in example: you can do Hours + Seconds or kilometers * miles, etc. It feels very organic and not like a "hack".

I prefer to encode quantities as types (and store value with most precision inside) and provide functions that return it in desired unit as int/long/whatever.

E.g. Duration type that stores nanoseconds and has to_seconds(), to_milliseconds() etc. It just feels more natural to me. Why should some function care which units are used? It just needs "duration" and will convert to desired unit internally (also it won't be part of its api which is good because it's unnecessary restriction).

Of course some C++ devs will disdain this approach because it's inefficient to pass highest precision value around when its not needed but for my use cases it doesn't matter.

Also, you should always use standard types when available. E.g. C++ has std::chrono, in Java world there are java.time types and kotlin.Duration.

That constant frustration is what made me write this post :)

I understand what you mean, and I even agree with it, but just to be a little pedantic, variable names are code, or at least they are more code than comments or docs.

But yes, encoding units into the type system is a much better solution. It doesn't work however for config options, environment variables or CLI switches.

I disagree. Every time I see duration: Long in code I want to find whoever wrote that and bash them on the head because what the fact this Long is? If unit is not immediately obvious in some way in current context (not necessarily as variable name), your code is not clear, period.

Related: Making Wrong Code Look Wrong

In that case I would call the variable fileSizeWithUnit and also document what the possible units are. I wouldn't say that documentation is categorically more important than good naming. Both are different aspects of good software development.

Then use bytes