README.md

metadata

license: cc0-1.0
pretty_name: Release tarball checksums (Debian, Homebrew, Guix)
tags:
  - file-hashes
configs:
  - config_name: data
    data_files: data/tarballs.parquet
  - config_name: maps
    data_files: maps/md5.parquet

Release tarball checksums (Debian, Homebrew, Guix)

Per-artifact checksums of release archives as shipped — the tarball itself, not its contents. Content-level indexes (Software Heritage, distro file digests) unpack archives before hashing, so a pristine hello-2.12.3.tar.gz matches nothing in them; the three package indexes here publish exact hashes of the artifact:

• debian — sid Sources.xz, all four components: md5 and sha256 per source file (orig/debian tarballs, .dsc, .asc) — each row is a single witness carrying both digests
• homebrew — formula.json: sha256 of each formula's stable upstream archive
• guix — sources.json, outputHashMode=flat entries only (recursive entries are NAR-style directory hashes, not raw bytes): sha256 of raw tarball bytes, with content-addressed mirror URLs

159,973 rows (debian 131,303 / guix 20,390 / homebrew 8,280), harvested 2026-08-11.

Layout

• data/tarballs.parquet — witness, sha256, md5, name, version, filename, loc; sorted by sha256. loc is the Debian pool directory, the Homebrew upstream URL, or the Guix content-addressed mirror URL.
• maps/md5.parquet — md5 → sha256 (Debian rows), sorted by md5.
• tarballs.sha256.bloom — DCSO-format membership filter over the 153,245 distinct sha256 keys (p = 10⁻⁴, 359 KiB, lowercase-hex keys). hdx filters fetch tarballs installs it so hdx scan can route matching files to this dataset.
• tarballs.tsv.gz — the normalized harvest the parquet is built from.

Both parquet files carry page indexes (zstd, no dictionaries, 128 KiB pages), so a point lookup over HTTP range reads fetches a few tens of KB. This is how hashdex resolves hashes from the CLI and from a static browser page.

Rebuild

release_tarballs.py in this repo fetches the three indexes (six metadata GETs, ~50 MB, no package content) and writes the TSV; tarballs_parquet.py converts and verifies (sorted keys, page indexes present). The bloom is built by hdx itself from the TSV's distinct sha256 column:

zcat tarballs.tsv.gz | cut -f2 | sort -u > sha256.txt
hdx filters build tarballs sha256 sha256.txt
# → ~/.cache/hashdex/filters/tarballs.sha256.bloom

Sources move — Debian sid continuously, Homebrew and Guix per commit — so a rebuild reflects harvest time; regenerate all three artifacts (parquet, maps, bloom) from the same TSV so they never disagree.

Checksums metadata © the respective projects (Debian archive, Homebrew, GNU Guix), generously public; this compilation CC0.