Learn

Erlang has two main structured-data options: records (older) and maps (modern, since R17).

Records are sugar for tagged tuples:

-record(person, {name, age, email}).

P = #person{name="Ada", age=36, email="[email protected]"},
P#person.name,                          %% access
P2 = P#person{age=37},                  %% update

Under the hood, P is the tuple {person, "Ada", 36, "[email protected]"}. The record syntax is just compile-time sugar.

Records have a fixed shape — once defined, you can't add fields without recompiling. They're fast and well-known but rigid.

Maps (#{...}) are dynamic key-value:

P = #{name => "Ada", age => 36, email => "[email protected]"},
maps:get(name, P),                       %% "Ada"
maps:get(role, P, undefined),            %% default if missing
P2 = P#{age => 37},                       %% update / add
maps:remove(email, P),                    %% remove key
maps:keys(P),                             %% list of keys

Pattern match on maps:

#{name := Name, age := Age} = P,
%% Name="Ada", Age=36

The := is the match-existing-key operator. Maps without that key would not match.

Maps vs records — when to use:

• Records — fixed-schema data, performance-sensitive, cross-module APIs
• Maps — dynamic data, JSON-like structures, optional fields, configuration

Most new Erlang code prefers maps for flexibility. Records still appear heavily in older codebases and OTP itself.