API Platform 4.3 — Pagination

Use when

• The collection can grow to millions of rows and COUNT(*) is too expensive.
• Infinite-scroll / autocomplete needs cursor pagination on a chronologically sortable key.
• Default 30 items per page is wrong for your domain (raise or lower it explicitly).
• Custom paginator needed (search engine, external API, computed projections).

Default workflow

1. Verify the 4.x defaults: pagination_items_per_page: 30, hydra_prefix: false. Adapt tests.
2. Cap paginationMaximumItemsPerPage (50-100) per operation — anti-DoS.
3. For large tables: paginationPartial: true (no COUNT, no totalItems).
4. For infinite scroll: identifier = UUID v7 / ULID + paginationViaCursor + ComparisonFilter on the cursor field.
5. For non-Doctrine sources: implement a custom PaginatorInterface.

Guardrails

• hydra_prefix: false is the default — tests must use member / totalItems / view / next (no hydra: prefix).
• Default page size is 30 in 4.x (changed from 20 in 3.x) — fix tests that hard-code 20.
• Always cap the max page size to prevent abusers from requesting ?itemsPerPage=10000.
• Cursor pagination requires a stable sort key — UUID v7 / ULID are ideal; auto-increment IDs work but leak volume.

Progressive disclosure

• SKILL.md lists posture and rules.
• reference.md carries the full patterns: 4.x JSON-LD shape, partial pagination, cursor with UUID v7, fetch-join collection / output walkers, custom parameter names, custom paginator, best practices.

Output contract

• Operations declare paginationItemsPerPage, paginationMaximumItemsPerPage, and where applicable paginationPartial / paginationViaCursor.
• Tests assert the 4.x JSON-LD shape (no hydra: prefix).
• Cursor pagination uses an indexed sort key; performance verified on a representative dataset.

References

• reference.md