Measure every declared primary and foreign key on the frozen tables

The release has always declared its keys (relationships.json, from core_relationships() and metadata/relationships_cross.csv) and gated a few of them — check_core_pk_unique() on the core primary keys, check_cruise_key_integrity() on the cruise_key edges — while the other ~50 declared foreign keys were true by construction and never measured. This walks the whole declaration on the assembled tables and writes the result as the integrity.json sidecar beside catalog.json, so a consumer (the schema browser, the docs) can show "unique, measured" and "0 orphans" per key instead of trusting the diagram.

Usage

check_release_relationships(
  con,
  rels,
  path = NULL,
  version = NULL,
  halt = TRUE
)

Arguments

con

DBI connection holding the frozen tables.

rels

Path to a relationships.json, or the list it parses to (primary_keys, foreign_keys).

path

Optional path to write integrity.json to.

version

Release version stamped into the sidecar (no wall clock: the file is deterministic for unchanged inputs, like the release stamp).

halt

stop() on any duplicate or NULL primary key, or any foreign-key orphan (default TRUE). A skipped key never halts.

Value

Invisibly, a list: primary_keys and foreign_keys data frames, ok, and n_skipped.

Details

For each primary key: rows, distinct key values, duplicates, rows with a NULL in any key column. For each foreign key: rows, rows whose key is NULL (permitted — a nullable edge is one where the source has nothing to point at, e.g. obs.taxon_key on an env row), and orphans: non-NULL values with no match in the referenced column. A key whose table or column is not in con is reported as skipped, never silently dropped.