inst/scripts/RELEASE.md

In taxadb: A High-Performance Taxonomic Database Interface

Publishing a taxadb snapshot

The data are built by this package, so a release is a package task rather than a separate pipeline. Everything below runs from a checkout.

1. Build

library(taxadb)
td_build(taxadb_providers(), version = "2026")

Downloads roughly 1.8 GB of provider archives on a cold cache (kept in build_dir() afterwards) and takes a few minutes. td_build() validates each table as it finishes and warns if any rule is violated.

2. Check

for (p in taxadb_providers())
  for (s in c("dwc", "common"))
    print(td_validate(p, s, version = "2026"))

Every rule must pass before publishing. td_validate() is the gate: a violation here means a reader will hit the same problem.

Point TAXADB_HOME at file.path(build_dir(), "out") so this reads the tables you just built rather than the published ones.

3. Write the metadata

td_write_metadata("2026")

Writes manifest.csv and README.md next to the Parquet files. The publish script refuses to upload without them.

4. Publish

export SOURCECOOP_KEY=...
export SOURCECOOP_SECRET=...
DRY_RUN=1 inst/scripts/publish-snapshot.sh 2026   # check the file list
inst/scripts/publish-snapshot.sh 2026

Publishing to an existing version overwrites it. People do read these URLs directly -- the advice on ropensci/taxadb#123 pointed at .../2026/dwc_col_part_0.parquet -- so overwriting changes data underneath running code. Prefer a new version unless the point is to replace something broken.

5. Re-render the README

README.Rmd queries the published data, so its output only reflects a new release after that release is up:

rmarkdown::render("README.Rmd")

Worth checking the diff: it is the quickest end-to-end confirmation that what you published is what you meant to.

Notes on individual providers

• gbif builds from the most recent backbone GBIF has published, dated 2023-08-28. Both hosted-datasets.gbif.org and ChecklistBank offer the same release; there is no newer one. Occurrence-derived taxon tables are not a substitute, since they carry no identifiers, no synonyms and no taxonomicStatus.
• col builds from latest_dwca.zip, so its content depends on the build date. The release it used is recorded in the manifest.
• ncbi's taxdump is regenerated daily and carries no version string; the date of the dump read is recorded as the version.
• fb and slb read FishBase's own published Parquet and download nothing. They are CC BY-NC, unlike the other providers.
• ott pins a release in build_ott(ott_version=); bump it when Open Tree publishes a new one.

taxadb documentation built on Aug. 31, 2026, 5:07 p.m.