API Platform 4.3 — Serialization

Use when

• Adding #[Groups] to a resource or DTO.
• Forcing relations to serialize as IRIs (or, exceptionally, embedded).
• Calculating fields without polluting the entity (custom normalizer).
• Adding admin-only or owner-only fields with a dynamic group.
• Tuning per-property date format / context with #[Context].

Default workflow

1. Define the read vs write groups (product:list, product:read, product:create, product:update).
2. Apply #[Groups] on every exposed property with a deliberate scope.
3. Force IRI on relations with #[ApiProperty(readableLink: false, writableLink: false)] unless embedding is explicitly required.
4. Add MaxDepth + enable_max_depth: true on circular relations.
5. For computed / role-dependent fields, decide: Context Builder (runtime), Mutator (build-time), or custom Normalizer.

Guardrails

• One group per direction. Never reuse product:read for writes.
• Sensitive fields out of every group. Use #[Ignore] for passwords, tokens, reset codes.
• BackedEnum for statuses. Never free string.
• hydra_prefix: false is the 4.x default — tests and clients consume member, totalItems, etc. without the hydra: prefix.
• skip_null_values: true is the 4.x default — no more "field": null in responses.

Progressive disclosure

• SKILL.md covers posture and decision tree.
• reference.md carries the full patterns: groups, MaxDepth, IRI forcing, BackedEnum, #[Context], custom Normalizer (ALREADY_CALLED pattern), conditional normalizer (Security-aware), Context Builder vs Mutator, gen_id, name converter, getValues() trick.

Output contract

• Resource / DTO with explicit per-property groups.
• Relations either IRI-only or embedded with a documented reason.
• Tests asserting the exact shape of the JSON-LD response.
• No null properties leaking in the response (verify skip_null_values).

References

• reference.md