You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Which parts of the indexer are generic and belong in arocapi, and which stay in ldacapi? Candidates to move upstream: deriveUniqueEntityIds, pickSingleMemberOf, entityAsCrate, building file storage meta, shared-entity upserts, a meta.crateId index, OpenSearch index settings and per-call refresh as OpenSearchBackend options. LDaC-specific parts (profile filter, search fields and facets, ldac: handling) stay here.
Decide what to propose upstream, in what order, and what ldacapi does in the meantime. Filing upstream needs the maintainer's sign-off first.
Decided by grilling. No ADR and no glossary change: the decision is cheap to reverse and matches what arocapi already does.
The boundary
arocapi stays read-only. ldacapi owns every write: entity and file rows, the crate↔entity links, and search documents. No crate-to-rows code is proposed upstream as part of this redesign.
The id rule and the file storage meta builder are plain functions with no LDaC imports, so they could move upstream if a second arocapi deployment ever needs them. No shared package and no profile-agnostic indexer core.
Input from "Oni aggregates: precompute or compute per request" (#73, resolved): aggregates are precomputed into a table of our own and read with one flat query per page from an entity transformer. arocapi never joins to that table. arocapi#142 (batch transformers) would replace the per-request microtask batching in src/oni.ts; nothing else is needed upstream for aggregates.
Activity
johnf commentedon Nov 17, 2026
Resolution
Decided by grilling. No ADR and no glossary change: the decision is cheap to reverse and matches what arocapi already does.
The boundary
Where each candidate ended up
meta.crateIdindex: not needed.meta.crateIdis dropped for our own link table (How shared entities are stored and deleted #72).pickSingleMemberOf: replaced by choosing the parent by LDaC profile (What the indexer derives from a crate #69). Ours.entityAsCrate: rewrites the virtual crate toldac:/pcdm:forms (What the indexer derives from a crate #69). Ours.deriveUniqueEntityIds(now the id rule) and buildingOcflStorageMetafor a file: generic, but a few lines each. They stay here.settingsoption for the index it creates crate-works/arocapi#143.indices.refreshonce when it finishes, through our own client.OpenSearchBackend.index: not proposed. We chunk our own calls, which we have to control anyway for large crates and file text.Upstream
settingsoption for the index it creates crate-works/arocapi#143 stays as it is; a pull request for it can be offered separately.In the meantime
settingsoption for the index it creates crate-works/arocapi#143 ships.Kept liftable
Handed on
johnf commentedon Nov 17, 2026
Input from "Oni aggregates: precompute or compute per request" (#73, resolved): aggregates are precomputed into a table of our own and read with one flat query per page from an entity transformer. arocapi never joins to that table. arocapi#142 (batch transformers) would replace the per-request microtask batching in
src/oni.ts; nothing else is needed upstream for aggregates.