[renovate]

This PR contains the following updates:

Package	Update	Change
iDvel/rime-ice	digest	99662ba → 927e1cf
jdx/mr-boxington	minor	v1.29.2 → v1.30.0
twpayne/chezmoi	patch	v2.74.0 → v2.74.1

---

Release Notes

jdx/mr-boxington

v1.30.0: : Configurable session history, richer cache-skip reporting, and blaze routing

Compare Source

mbx analyze and mbx stats now explain why compilations skipped the cache and where private incremental state came from, and you can configure how much session history to keep. This release also fixes several cases where caches or target directories were evicted in the wrong order, and stops terminal and piped builds of workspace crates from keeping separate cache entries.

Added

• mbx analyze and mbx stats explain uncached compilations (#​911, @​jbellis). Session history now records why a compilation skipped the cache lookup, which private crates it links, and where its private incremental state came from. When a crate rebuilds only because it links another crate's private output, mbx analyze now charges that time to the crate that changed instead of calling it a "first build". A new section shows where private incremental state came from: seeded from another checkout (with bytes copied and time taken), this checkout's earlier state, started empty (with a reason), or seeding failed. mbx stats adds lifetime totals by outcome, for example 7 hits, 2 misses, 3 not looked up, 1 bypassed since 2026-11-05, and mbx stats --json adds savings.compilations. These totals count only builds recorded by this version or later, and builds from older versions are reported as before.

• Configurable session history retention (#​912, @​jbellis). mbx explain --last, mbx analyze and mbx tui read session history, which was always capped at 256 sessions and 7 days. On machines that run many parallel builds, that could cover only a few hours. Both limits can now be changed, and the defaults stay the same:

  [gc]
  events_max_age = "7d"       # "none" keeps sessions of any age
  events_max_sessions = 256   # "none" keeps any number of sessions

  You can also set them with MBX_GC_EVENTS_MAX_AGE and MBX_GC_EVENTS_MAX_SESSIONS. A session is removed when it reaches either limit, but never while a build is still writing to it. Higher limits use more disk space, since each session file can grow up to events_max_size (16 MiB by default).

Changed

• MBX_BLAZE=1 runs blaze when it is installed (#​904, @​jdx). Current blaze releases no longer include ra-jit, so this setting did nothing and printed a warning about a missing ra-jit. The Cargo shim now sends dev build, run, test and check commands (and their b/r/t/c aliases) to blaze if it is on PATH. A routed cargo test gets BLAZE_TESTS=all unless you set it yourself, so every test runs, as with Cargo. If only ra-jit is installed, mbx still runs ra-jit cargo … as before. If neither is installed, mbx warns and builds with Cargo. Note: blaze never calls Cargo, so mbx's cache doesn't see builds that blaze runs. If you have both programs installed, blaze now runs instead of ra-jit. Flags that blaze does not support make the build fail instead of falling back to Cargo.

Deprecated

• The ra-jit fallback for MBX_BLAZE will be removed in mbx 2.0 (#​904).

Fixed

• Store eviction tracks real use on macOS, Windows and noatime mounts (#​903, @​ViRb3). On APFS, on Windows volumes where last-access updates are off (the default above 128 GB), and on Linux noatime mounts, reading a file does not update its access time. As a result, eviction removed the oldest-written objects first, even ones that every build restores. Now mbx records the access time itself, at most once a day per file, whenever a build restores a cached result or mbx cache import restores workspace state. The modification time is not changed. Lookups that don't restore anything, such as prefetch, export, gc and mbx cache verify, don't count as use. mbx doctor now checks that mbx can record use times (ok atime …: mbx records use times itself) and no longer warns on every Mac.
• mbx gc keeps target directories whose programs you run between builds (#​905, @​jdx). Some programs are built rarely but run often from target/, such as a hook binary or one linked from ~/.local/bin. Their target directories were ranked by last build only, so they were removed first under target.max_size or expired under target.max_age. Collection now uses whichever is later: the last build, or the last time a top-level executable or library in a profile directory (such as target/debug/app) was accessed. A target that is still being run but not rebuilt also keeps its build units. Limitations: this only works on filesystems that track access times. Under Linux relatime, a run counts at most once a day. Hard-linked cache restores don't count. When the disk is below gc.min_free_size, these targets can still be removed.
• Terminal and piped builds share workspace crate cache entries (#​907, @​jdx). Cargo passes --diagnostic-width only when stderr is a terminal, so builds in terminals of different widths and piped builds (editors, agents, CI) kept separate entries for workspace crates and kept rebuilding each other. The flag no longer affects the cache key for any crate. Workspace crate entries cached with a width will miss once after the upgrade. Diagnostics replayed from the cache keep the line wrapping of the build that produced them.
• Edited workspace crates stay incremental with nightly Cargo (#​909, @​jdx). Nightly Cargo (1.101) empties a unit's build/<package>/<hash>/out directory before recompiling it. That removed the dep-info mbx uses to detect edits, so every edit reported 1 misses, 1 incremental instead of moving to private incremental state. mbx now keeps a copy of the dep-info and restores it when the directory is empty. Builds with stable Cargo are unaffected.
• mbx explain --last and mbx analyze work in nested checkouts (#​910, @​jbellis). In a checkout inside another Cargo project's directory, such as agent worktrees under <repo>/.mj/lanes/<name>, both commands looked up the outer project's builds and failed with no recorded build was found. They now use the deepest recorded workspace that contains the current directory and print a workspace: line showing which one they read.
• cargo run -- -h stays under mbx (#​918, @​joshka). mbx treated -h or --help after -- as a request for Cargo's help. The program then ran without mbx and built into ./target, recompiling dependencies. Arguments after -- now go to the program. cargo run -h and cargo help run still show Cargo's help.
• Fewer stale progress rows after resizing the terminal (#​922, @​jdx). When the terminal got shorter, the top of the pretty progress block scrolled into scrollback, where mbx can't erase it, and left behind duplicate rows. On tall terminals the block is now limited to half the terminal height, with a minimum of 14 rows. Shrinking the window to just over half its height no longer leaves rows behind. Shrinking it further leaves fewer rows than before.

New Contributors

• @​joshka made their first contribution in #​918

Full Changelog: jdx/mr-boxington@v1.29.2...v1.30.0

twpayne/chezmoi

v2.74.1

Compare Source

Changelog

Fixes

• 2cae8ab fix: Make chezmoi init --apply really run chezmoi apply
• 27c52c8 fix: Ensure config template data is reloaded in chezmoi init --apply
• 595c769 fix: Fix WinGet config read in Windows upgrade command

Documentation

• effc262 docs: Add link to video
• 7c93096 docs: Add link to github.com/dirichy/vimoi.nvim

---

Configuration

📅 Schedule: (in timezone Asia/Singapore)

• Branch creation
  • Between 12:00 AM and 03:59 AM (* 0-3 * * *)
• Automerge
  • At any time (no schedule defined)

🚦 Automerge: Enabled.

♻ Rebasing: Whenever PR is behind base branch, or you tick the rebase/retry checkbox.

👻 Immortal: This PR will be recreated if closed unmerged. Get config help if that's undesired.

---

• If you want to rebase/retry this PR, check this box

---

This PR was generated by Mend Renovate. View the repository job log.