Connect arbitrary source URLs to canonically-named upstream projects.
Build recipes (like Julia's JLLs) download sources from all over: project websites, mirrors, GitHub archives, git commits. Nothing in those URLs says which upstream project or release they are, yet that's the first thing you need to match a package against security advisories or check whether it's out of date. Repology already gives thousands of projects a single canonical name and tracks the URLs, repositories and CPEs distributions use for them. Repology.jl uses a snapshot of that data, plus hand-curated metadata, to map a source URL (and optional commit) to a Repology project and upstream version, and records how each answer was found.
using Pkg
Pkg.add(url="https://github.com/JuliaPackaging/Repology.jl")using Repology
id = Repology.identify("https://github.com/libexpat/libexpat/releases/download/R_2_8_3/expat-2.8.3.tar.xz")
(id.project, id.version, id.provenance)
# ("expat", "2.8.3", "url-pattern")Names are qualified (the module exports none). An unknown URL returns an
identification with nothing fields; a known project can lack a version. Pass
commit= for repository sources; tag lookup may access the network and clone
the repository. See the guide and API reference.
additional_info.toml supplies per-project URL patterns, repositories, tag
patterns, hand-verified sources, CPE prefixes, and version schemes. Add regression
tests with curation changes and run julia --project=. -e 'using Pkg; Pkg.test()'.
Metadata is cached, so restart Julia after edits.
repositories.toml, downloads.toml, and cpes.toml live in the repology_data
artifact bound by Artifacts.toml. Repology.data_dir() resolves its directory;
set REPOLOGY_DATA before loading data to test an alternate snapshot.
The weekly/manual Export from Repology's DB dump workflow extracts a refresh
with scripts/update_repology_info.jl, retaining older entries absent from the
new dump. It tests the corpus and packages/uploads changed data with
scripts/publish_data.jl into immutable data-YYYY-MM-DD releases (plus a run id
for repeat refreshes). Clean refreshes push the artifact binding; regressions or
changed project mappings produce a review PR. Never delete or replace a data
release referenced by any package version. The
guide covers local refreshes and curation.