[windows_sucks]

    make_pepperoni_pizza()
     is bad code compared to
     make_pizza(toppings=[PEPPERONI])

How would you make Hawaiian pizza? I forget, does it include Ham? or just pineapple? you're forced to the remember that nuance in your suggested implementation, but not with "make_hawaiian_pizza()"

[frognumber]

It's data.

   make_pizza(toppings=HAWAIIAN_TOPPINGS)

or

   make_pizza(HAWAIIAN)

or similar. Data should generally not be hard-coded, both because it changes and because it wants to be validated. Starting with:

   HAWAIIAN = { TOPPINGS: [ PINEAPPLE  ...

is okay. That can later be loaded from a config file, a database, or otherwise, as the system expands.

[HelloNurse]

There's an interesting architectural decision here: what form of the pizza recipes database strikes the right balance between too hardcoded and too complex. I'd use some kind of configuration file or RDBMS, constants are more readable but still out of place as part of code.

[frognumber]

The nice thing about constants is that you can't make typos. A string like "peperoni" isn't caught, but toppings.PEPERONI will fail immediately.

I use Python. It's easy enough to, for example, make an `enum` from entries in a config file or even a database:

https://docs.python.org/3/library/enum.html

Scroll down to the functional API. There are many similar design patterns. The nice thing about these is that you get automatic type checking. If your config file is:

    toppings: ['pepperoni', 'ham', 'pineapple'],
    pizzas: {
       'Hawaiian': ['pinapple']
    }

If you load this in as strings, it will load and later silently fail. If you add validation code, you'll only validate what you remember. If you make an enum or similar custom type, it necessarily will fail-on-load, and probably with a reasonable error.

The major downside -- which is really incidental (due to poor library design) -- is that most JSON/YAML libraries won't reasonably serialize/deserialize non-Python types. So there's a bit of (unnecessary) overhead there.