Skip to content

CHANGELOG.md

Latest commit

 

History

History
173 lines (105 loc) · 12 KB

CHANGELOG.md

File metadata and controls

173 lines (105 loc) · 12 KB
Raw

Changelog

All notable changes to cargo-reclaim will be documented in this file.

Format follows Keep a Changelog. cargo-reclaim uses Semantic Versioning.

[0.5.0] - 2026-07-22

Added

  • Background triggers are now an array — configure any number of independent [[background.trigger]] blocks, each with its own cadence, limiter, policy, and disruptiveness. A trigger with no limiter is periodic; one with a limiter fires only when breached. (Replaces the single [background.periodic] / [background.trigger] tables.)
  • Per-trigger whole_target override (defaults to [policy].whole_target). Unattended delete still requires the aggressive policy and allow_unattended_whole_target_delete = true.
  • Per-trigger disruptiveness toward active builds:
    • interrupt_active_build = true — delete in-use artifacts and whole targets even while a build runs (the build fails when its files vanish). Off by default; a normal trigger protects active builds entirely.
    • kill_active_builds = true — before cleaning, terminate the cargo/rustc processes building targets under the config roots (SIGTERM, a 5-second grace, then SIGKILL), so the disk fill stops and there is no active build left to protect. Only build processes inside roots are ever signalled, and never cargo-reclaim itself. This makes an emergency only_when_disk_free_below/whole_target = "delete" trigger able to stop a runaway build and reset its target instead of letting the disk reach 100%.

Changed

  • The single-table [background.periodic] / [background.trigger] form (0.3–0.4) is replaced by the [[background.trigger]] array. The deprecated flat mode/check_every form still normalizes into one trigger.
  • Config parsing is now strict: an unknown or misspelled key in any config table is rejected with an actionable error at load time instead of being silently ignored. A wrong key can no longer leave a trigger mis-limited or a setting quietly unset.

Fixed

  • A [[background.trigger]] limiter written as only_when_disk_free_below (the documented spelling, matching the flat [background] form and the README) was silently ignored, because the array parser only accepted a different key name. The limiter therefore read as empty, so a disk-gated sweep/emergency trigger fired on every cadence regardless of free space — killing builds and cargo-cleaning targets when the disk was nowhere near the threshold. The disk-free limiter key is now only_when_disk_free_below everywhere, and unknown keys fail loudly (see above), so this class of silent misconfiguration cannot recur.
  • Active-build protection no longer depends solely on the point-in-time process scan, which could be sampled in a gap between rustc invocations (or miss a build driver such as cargo-nextest it does not recognize) and let StaleDeps/StaleIncremental artifacts be deleted mid-build — deleting a live --all-features feature-variant the running build still links, which cargo will not rebuild while its fingerprint DB considers it fresh, breaking the build. A build writes into its target continuously, so the target's newest artifact mtime is now used as a race-free signal: while any artifact in a target was written within [planner].recent_write_keep_window, the whole target is protected — including the stale classes, whose own mtimes are old by definition and were previously guarded by the process scan alone. Reclaim still happens between builds, and a disruptive interrupt_active_build/kill_active_builds trigger still opts out.

[0.4.0] - 2026-07-22

Added

  • New sweep policy: cargo-sweep-style reclamation of cold final binaries (final_executable, final_rlib, final_library, final_wasm) once they are older than a sweep age threshold, on top of the balanced removable set. It never deletes whole targets, docs, packages, or unknown files, and — like every policy — reclaims nothing from a target with an active build. Configure the age gate with [planner].sweep_older_than (default 24h).
  • Per-trigger policy override: a [background.periodic] or [background.trigger] block may set its own policy (e.g. policy = "sweep"), so a disk-pressure trigger can reclaim more aggressively than the routine cadence while still satisfying the unattended high-policy gate.

Notes

  • Active builds remain fully protected: while any cargo/rustc process is touching a target, cargo-reclaim reclaims nothing from it (it cannot distinguish a superseded hash variant from a live feature-variant the linker needs without cargo's fingerprint DB). Age-based reclaim happens between builds, where cargo re-plans and rebuilds anything removed.

[0.3.0] - 2026-07-21

Added

  • Split the [background] watcher into independent, composable trigger blocks: [background.periodic] (fires on a timer) and [background.trigger] (fires on a poll). Configure either or both, so a routine cadence and a responsive disk-pressure gate can run at once — the latter can trim the instant free space crosses a threshold instead of waiting for the next periodic pass.
  • Introduced the limiter concept, orthogonal to the trigger: each block may carry only_when_disk_free_below, min_free_disk, or max_target_size. With no limiter a fired run always cleans; with a limiter it cleans only when a threshold is breached. Disk limiters use a cheap free-space check; max_target_size scans target sizes.
  • Surfaced non-fatal config deprecation notices through ReclaimConfig::deprecations, printed as warnings by the scheduler commands.

Changed

  • mode is no longer a [background] key; how a run is triggered is now expressed by the presence of the periodic/trigger blocks. Policy and budget config still govern what a run removes and how much.

Deprecated

  • The flat [background] keys mode, check_every, only_when_disk_free_below, and min_free_disk are still accepted and normalized into the new blocks (with a warning), and will be removed in 0.4. mode = "periodic" maps to a [background.periodic] block; mode = "threshold" maps to a [background.trigger] block that inherits [policy].max_target_size as a limiter.

Fixed

  • Kept the background inventory from aborting a run when a concurrent cargo build deletes an artifact (for example a deps/*.rcgu.o object file) between directory enumeration and the snapshot's stat; the vanished path is now skipped like the stale-deps and stale-incremental passes already did, instead of failing the run.

0.2.2 - 2026-07-07

Fixed

  • Rejected a plan --json dry-run report (or an unrecognized file) passed to apply --plan, edit-plan --plan, or cargo-home apply --plan with an actionable error that points at --save-plan, instead of a raw serialization error about a missing id field (#1).
  • Discovered a shared CARGO_TARGET_DIR or build.target-dir that lives on a different filesystem than the project root without requiring --cross-filesystems; the flag now governs incidental traversal only, not explicitly configured output locations (#2).
  • Recognized a cargo target directory by its .rustc_info.json marker regardless of directory name, so a shared target directory named e.g. cargo-target is listed and cleanable; the generic CACHEDIR.TAG marker still requires the conventional target name (#2).

Changed

  • Explained empty list results, distinguishing "no Rust project found under the scanned roots" from "Rust projects found, but no cleanable target directories," in both terminal and JSON output.

0.2.1 - 2026-07-05

Fixed

  • Summarized foreground cleanup/apply terminal output by default and wrote complete per-run JSON reports under the cargo-reclaim state directory.

0.2.0 - 2026-07-05

Changed

  • Made cargo-reclaim <roots...> the primary cleanup assistant entrypoint, with smart trim as the default cleanup mode for active projects.
  • Made cargo-reclaim list <roots...> the read-only target inventory surface and removed the old public targets command surface.
  • Moved explicit whole-target deletion to the cleanup assistant path with --target <path> --delete-target --yes, keeping whole-target cleanup separate from default smart trim.
  • Updated CLI help, JSON inventory output, README examples, and release preparation metadata around the 0.2.0 command model.

Added

  • Real terminal assistant coverage for selector, mode, confirmation, cancellation, page-skipping flags, and non-TTY guard behavior.
  • Deterministic CLI integration test isolation from live cargo/rustc process scans.

Fixed

  • Improved target inventory sizing throughput and stale plan-test stability before the 0.2.0 release.

0.1.1 - 2026-07-04

Fixed

  • Support Cargo subcommand invocation through cargo reclaim ... by accepting Cargo's leading reclaim shim argument before normal command parsing.

0.1.0 - 2026-07-03

First release of cargo-reclaim: safe Cargo artifact cleanup for real Rust workstations.

Added

Planning and apply

  • Read-only scan and plan workflows for Cargo target directories.
  • Persisted cleanup plans with expiration and apply-time revalidation.
  • Policy modes: observe, conservative, balanced, aggressive, and custom.
  • Ignore, skip, recent-write, keep-size, and rustc/toolchain preservation controls.

Target cleanup

  • Cargo target discovery from project roots, Cargo configuration, and target-root evidence.
  • Interactive target directory discovery and selected whole-target cleanup through cargo-reclaim targets and cargo-reclaim targets clean.
  • Partial cleanup classes for incremental artifacts, fingerprints, build-script caches, dep-info files, temporary files, object metadata, stale fingerprint-group intermediates, stale hashed deps variants, old deps outputs guarded by recent-write windows, and stale incremental sessions or unit variants.
  • Protected classes for whole targets, docs, packages, timings, final binaries, final libraries, final .rlib, final .wasm, and unknown artifacts.

Cargo home

  • Cargo home cache reporting.
  • Cargo home cleanup plans and saved-plan apply.

Scheduler

  • Config-driven scheduler execution.
  • Resident scheduler service support with platform preview/install/uninstall flows for Linux systemd-user, macOS launchd, and Windows Task Scheduler.
  • Active cargo/rustc process detection on Linux through procfs and on macOS/Windows through a native process-table provider.
  • Threshold background cleanup with per-target size ceilings, target size goals, absolute free-space triggers, and durable service state.
  • Durable scheduler run logs and state directories.

Cargo config

  • Read-only Cargo config recommendations.
  • Cargo config preview and explicit apply flow.

Documentation and automation

  • Saved plan editing with list, selector, and interactive modes.
  • JSON output for automation-friendly scan, plan, scheduler, target, and Cargo home workflows.
  • Deep directory measurement for CLI, scheduler, and background service planning, so cleanup budgets and reported sizes reflect real directory contents.
  • Scheduler service status that keeps running when PID liveness cannot be inspected from the current environment instead of falsely reporting a stale service.

Safety

  • Destructive flows require explicit confirmation with --yes.
  • Saved apply flows revalidate path kind, size, modification time, and symlink state before deletion.
  • Whole-target deletion remains separate from partial artifact trimming and requires explicit selected target cleanup or whole-target policy configuration.