One row per (dataset, finding); a clean dataset has a single ok row. The
structural half always runs (catalog_findings()); the network half
(network = TRUE, i.e. not CALCOFI_SKIP_LINK_CHECK) probes every
distribution URL once with a one-byte ranged GET — 404/410/451 is url_dead
(error), 5xx / no answer is url_unreachable (warn); retired and
superseded rows are not probed (they are expected to be gone). Holdings
are checked for name, category, provider and their links.
Usage
check_dataset_catalog(
record,
registries = NULL,
network = TRUE,
probe = NULL,
timeout = 30,
pending_base = NULL
)Arguments
- record
from
build_dataset_catalog()(or adatasets.jsonpath)- registries
from
read_catalog_registries(); NULL trusts the record's ownregisteredflags- network
probe the URLs (default TRUE)
- probe
the probe function
function(url) status; the tests inject one- timeout
seconds per request
- pending_base
a URL root this release writes after the check (its STAC root,
release_stac_base()); a dead URL under it isurl_pending. NULL: none.
Value
A tibble: dataset_key, finding, level,
detail, url, exempt, question.
Details
A URL this release has not written yet is url_pending, not url_dead. The record lists
each public dataset's STAC collection under the release's own STAC root, and
release_database.qmd uploads that tree AFTER this check. A dataset in its first release has no
collection there yet, so the check failed every new dataset (first seen with
calcofi_ctd-derived, 2026-09-24, the first new dataset since STAC landed on 2026-09-05), while
the existing ones passed only on the previous run's objects. With pending_base, a
404/410/451 under it is url_pending (level pending, never blocking here), and
recheck_pending_urls() re-probes exactly those rows once the tree is uploaded, turning any
that still fails into url_dead.