The difference turns on the meaning of identity, which philosophers have long pondered and debated, especially in regard to individual humans (i.e. the question "Who am I really?"). The answers fall into two rough categories: intrinsic and extrinsic. Unsurprisingly, the dividing-line can be fuzzy.
- In the extrinsic category, a thing's identity is tied to sensations and ideas about that thing. This would be like identifying my phone by its shape, color, model, and so on. Based on extrinsic identity, if any of these "extrinsic" propositions about my phone ceased to be true, my phone would therefore no longer have the same identity. The phone would not only appear but be a different phone. Its identity is dependent. When someone meets a new acquaintance, his or her identity is generally rather extrinsic. "I don't know this person, but his name is Fred and he works in Human Resources."
- The opposite of extrinsic identity is intrinsic identity, which is independent of sensations and ideas about a thing. A thing with intrinsic identity "wears" sensations and ideas but is "itself" regardless. This would be like disassembling my precious phone piece by piece and then continuing to call the pile of pieces my phone. Transformations of all kinds don't change intrinsic identity. When someone has maintained a long-term connection to another person, his or her identity is generally rather intrinsic. "She's changed her residence, weight, politics, and career over the last several years, but she's still Betty to me."
- If the object's identity consists of a particular combination of attributes, then its identity is extrinsic. Two objects of the class with identical attribute values are indistinguishable (for the purpose of the domain). These are known as value objects. Each instance is a "value" similar to a number or a string of characters. Since extrinsic identity is by definition fully dependent on a thing's parts, a value object never changes. But someone often creates a new value object using the attribute values of preexisting value objects. This is just like adding two numbers to yield a third number. Modelers and developers can exploit the unalterable nature of value objects for various advantages, such as avoiding the need for storage. Possibly, there could be a highly reusable "mathematics" of the value object in which all methods are "operations" that 1) operate on and return value objects and 2) are completely stateless (free of side effects).
- Instead, a domain object's identity could be persistent throughout arbitrary transformations, which indicates intrinsic identity. Instances of the class probably have vital "history" and might not represent the same physical domain item despite identical or near-identical attribute values. These are known as entities. Normally a nontrivial entity will aggregate one or more value objects but along with a mechanism of unique identification. That mechanism could be a GUID or something else as long as no two instances can contain the same identifier by accident. Unlike value objects, (long-lived) entities change and correspond more directly to physical domain items, so a storage strategy for the entity and its associated value objects is necessary. (The advice of DDD is to keep the entities' code as purely domain-driven as possible by abstracting the storage details in a repository object.)