Manage the machine cache

Everything reusable lives in one cache, shared by every repository: ${XDG_CACHE_HOME:-~/.cache}/prec/, or PREC_CACHE. Outputs and scratch stay in each repository’s tmp/.

folder	holds	keyed by
src/<repo>/<commit>/	that commit’s source, checked out by git read-tree with a private index, read-only ({app})	repo and commit
work/<repo>/	the build tree, moved between commits so only changed files change	repo
target/<repo>/	the repo’s shared cargo target dir, so a new commit rebuilds incrementally	repo
bin/<repo>/<commit>-<build>/	the built binaries ({bin})	repo, commit and the build command’s hash
locks/	one lock per entry, so a prune never removes an entry being built or used
archive-queue/	renders waiting for an archive that was offline

A <repo> key is the repository’s name and a hash of its origin URL, so every checkout and worktree of one repository shares entries.

See it

prec cache

cache: tmp/cache (0 B)
       set by PREC_CACHE
  (empty)

With entries, prec cache lists each with its size and last use. prec cache --json prints one JSON object, specified in Formats; it never waits on a lock.

Prune it

prec cache prune                    # entries unused for 30 days
prec cache prune --older-than 2w    # or 12h, 90m, 30d
prec cache prune --all              # every entry not in use
prec cache prune --json --dry-run   # what a prune would remove, removing nothing

{"freed":0,"kept_in_use":0,"removed":[],"v":1}

An entry a render holds stays, and the JSON’s kept_in_use counts them. Removing an entry costs only time: the next render of that commit checks it out and builds it again.

prec downloads nothing, so the cache only ever holds your own repositories’ sources and builds.