Changelog

radicle-httpd + radicle-search 0.27.0

Commit-count endpoint: New /repos/{rid}/stats/commits/{sha} counts reachable commits from the commit-graph, avoiding the full history walk of /stats/tree
Diff-stats endpoint: New /repos/{rid}/diff/{base}/{oid}/stats returns line counts without building hunks, 6–15x faster than /diff; it skips rename detection, so counts can differ slightly on renamed or copied files
Faster file browsing: Blob requests no longer walk the whole history to find a file's last commit, which took seconds to minutes on large repositories
Repository reads off the async workers: Synchronous libgit2 and COB reads run on the blocking pool, so one slow read no longer stalls queued requests
Repo aliases in the JSON and raw APIs: --alias path segments resolve there too, not just on the git routes, and the repo endpoint exposes an optional alias field
Higher open-file limit: Raised at startup, fixing spurious "missing object" errors in storage with thousands of packfiles
Empty job COBs excluded: Job listings omit jobs with no runs
RUST_LOG in the default logger: Honoured without the logfmt feature, defaulting to info
Updated radicle crates, cargo dependencies and the Rust toolchain (1.97.1)

Repository search: New optional radicle-search indexing daemon backs a /repos/search?q=… endpoint for typo-tolerant search over repo names and descriptions, plus faster /repos?sort=activity|seeding listings; httpd falls back to its built-in storage walk when the daemon is absent, so existing behavior is unchanged
Git archive downloads: Additional formats ZIP and TAR are now supported. Also, a prefix is added to the archive by default, which can be disabled by using the query ?prefix=false. This query string is immediately deprecated. It only exists as a workaround for users to adapt their download scripts etc. to accommodate for the prefix. It will be removed in a future version.
Streaming archive downloads: Archives are streamed to the client instead of being buffered in memory, avoiding excessive memory use on large repositories
Faster archive HEAD requests: HEAD requests no longer build an archive that is then discarded
Streaming git operations: Clone, fetch, and push stream request and response bodies and accept gzip-compressed requests, reducing memory use on large operations

Canonical and peer refs in repo API: The repo endpoint now exposes canonical refs and per-peer refs, letting clients discover branches and tags resolved by delegate quorum as well as those from individual nodes
Job COBs endpoint: New /repos/{rid}/jobs/{sha} endpoint serves job collaborative objects, enabling clients to query CI/CD job state

Windows support: The HTTP daemon now runs on Windows with platform-specific adaptations
Faster API responses: Improved /node endpoint performance
Git namespace paths: Direct access to node ref namespaces via /{rid}.git/{nid}/ remote URLs, simplifying clones and fetches from specific nodes
Updated dependencies and internal improvements

Live config reload: Update configuration without restarting by sending a SIGHUP signal