Union types deduplicate identical members while sum types preserve distinct tags

The core distinction between union and sum types lies in how they handle the set of possible values. A union type is mathematically defined as the union of its member sets, which means if two members have identical value spaces, they collapse into a single type. For example, a union of two bool types simplifies to just bool, because the set of possible values remains {false, true}. This behavior makes unions suitable for expressing uncertainty or combining overlapping type spaces, where the specific origin of a value is irrelevant.

Sum types, also known as tagged unions, solve the problem of conflation by wrapping each member in a distinct tag. If you need to distinguish between a Gregorian year and a Hijri year, both stored as u16, a simple union would collapse them into a single u16, losing the semantic distinction. By creating separate structs for each variant, the type system forces the program to explicitly handle the tag, ensuring that data is interpreted correctly despite identical underlying representations. This adds a layer of safety for domains where structural similarity does not imply semantic equivalence.

The practical implication for engineers is that union types facilitate type inference and simplification in type systems that support set operations. If a function can return either u16 or i16, and another returns u16 or u32, the combined return type of calling either function simplifies to the union of all three types. This automatic deduplication reduces the cognitive load of tracking type possibilities. However, this same mechanism is a liability when distinct concepts share a representation, requiring the developer to switch to sum types to maintain data integrity.

Adopting sum types over unions incurs a cost in verbosity, as each variant must be explicitly defined and tagged. This is a necessary trade-off for correctness in complex data models, but it can feel redundant for simple cases where the distinction is trivial. The decision to use one over the other should be driven by whether the data's meaning depends on its origin or only on its value. If the origin matters, sum types are mandatory; if it does not, unions provide a more concise and mathematically consistent model.