What are dunder methods and how do they fit into the data model?

Dunder methods are the special methods with double underscores that Python syntax dispatches to. Calling len on an object invokes dunder len, the plus operator invokes dunder add, and a for loop invokes dunder iter. Implementing them makes your class behave like a built in type, so the language protocols work on your objects instead of you inventing custom method names.

They are the hooks the interpreter calls when you use syntax. Square bracket access goes to dunder getitem, the in keyword goes to dunder contains, equality goes to dunder eq, and so on. The point is that Python defines behavior as protocols rather than interfaces you inherit, so if my object implements dunder iter and dunder next it is an iterator, full stop, no base class required. In practice the ones I implement most are repr, eq, and hash. Repr especially, because a good repr showing the identifying fields turns a useless debugging session into a five minute one, whereas the default object at 0x7f prints tell you nothing. I am careful with eq and hash together, since defining eq without hash makes instances unhashable and they silently stop working as dict keys. On a domain model I built, adding dunder lt let us sort records with the built in sorted call instead of passing a key function everywhere.