grit://cooking/changelog← what’s cooking
Sunday, November 8, 2026 · release v0.5.32

17 changes landed on main

5 roadmap steps across 2 items · 5 issue fixes · 7 other changes

v0.5.32 moves most of the library API onto Repository handles and options structs, and fixes memory use and output bugs in fsck, checkout, status and clone.

  • ›Branches, tags, notes, checkout, status, staging, commit and hooks now live on Repository handles; legacy free functions removed
  • ›Text diffs go through one UnifiedDiffOptions; tag listing is Tags::list(&TagListOptions)
  • ›fsck verifies large blobs by streaming; 1 GB blob no longer needs ~1 GB RSS (#2564)
  • ›Large checkout/restore under autocrlf streams instead of double-buffering, with a CRLF chunk-boundary fix (#2565)
  • ›Windows status shares FindFirstFileExW listings, targeting the 130 ms vs 47 ms gap on cargo (#2566)
  • ›Clone stores an absolute local origin URL (#2562); --stat bar aligns with CJK/emoji names (#2567)
$ git log 90f552da..v0.5.3249 commits on GitHub ↗300+ files +11,271 −6,290
Highlightsmeasured by the reviewers, before → after
Windows (autocrlf=true): restoring a 1.5 GB file peaks at 3.1 GB RSS (git: 1.5 GB); autocrlf=false streams
256 MiB autocrlf restore peak RSS11× better████████████████████████ 572,154 KiB ██ 50,062 KiB
Text diff rendering behind one options struct
public too-many-arguments allowances3.5× better████████████████████████ 7 functions ███████ 2 functions
Close-out: stragglers, docs and the 75% ratchet
Public too-many-arguments allowances3.5× better████████████████████████ 7 functions ███████ 2 functions
Branches, tags, notes, switch, checkout, reset, restore, clean, mv, rm on handles
public repository-first free functions24% better████████████████████████ 175 ratchet count ███████████████████ 141 ratchet count
Status, staging and commit on handles
add/modified@synth-10k grit median (git 16.391 ms)4% better████████████████████████ 10.5 ms ███████████████████████ 10.0 ms
roadmap · Collapse function variant families into options structs and handle methods4 steps
  • ›Branches, tags, notes, switch, checkout, reset, restore, clean, mv, rm on handlese57903e4

    Moves branch, tag, notes, checkout/switch/reset/restore/clean/path and workspace-snapshot operations behind repository-owned handles, then rewires the CLI and tests to use those supported entry points. The round-two fixes close the remaining public API leaks and collapse tag listing to `Tags::list(&TagListOptions)`.

    public repository-first free functions: 175 ratchet count → 141 ratchet count (24% better)
  • ›Status, staging and commit on handlesd450ed5c

    Moves status, staging, hunk selection, and commit planning/creation behind repository-owned handles, leaving the CLI as thin forwarding code. Round four adds executable module examples for the newly canonical `commits` and `status` modules and updates remaining docs/call sites to the handle API.

    add/all@synth-100k grit median (git 143.717 ms): 113.8 ms → 113.8 ms (0% worse)add/modified@synth-10k grit median (git 16.391 ms): 10.5 ms → 10.0 ms (4% better)commit/small@small grit median (git 5.498 ms): 5.0 ms → 5.0 ms (1% worse)
  • ›Text diff rendering behind one options struct274e6266

    This consolidates text patch rendering and line-change counting behind one typed `UnifiedDiffOptions` API, and routes combined patch formatting through typed options as well. It keeps the computations as library free functions, removes dependency types and positional boolean-heavy signatures from the public surface, and preserves Git-compatible output across algorithms, hunk controls, function matching, anchors, whitespace modes, labels, and EOF markers.

    userdiff rust_match_line Criterion median: 278.2 ns → 276.2 ns (1% better)public too-many-arguments allowances: 7 functions → 2 functions (3.5× better)public functions with two-plus bare bools: 16 functions → 10 functions (60% better)
  • ›Close-out: stragglers, docs and the 75% ratcheta6e86a2c

    Consolidates the pre-0.6 library surface around Repository-owned handles for hooks, worktrees, blame, references and related operations, while removing legacy aliases and free-function variants. Its stacked dependency also replaces the text-diff function family with typed options and shared rendering, with CLI, examples, tests, docs, snapshots and ratchets updated together.

    Public functions containing `_with_`: 28 functions → 24 functions (17% better)Public too-many-arguments allowances: 7 functions → 2 functions (3.5× better)Public functions with 2+ bare booleans: 16 functions → 5 functions (3.2× better)Repository-first public free functions: 52 functions → 28 functions (86% better)
roadmap · Pluggable object store on Repository1 step
  • ›Before/after benchmark sign-offf49de4d8

    This branch records before/after daily headline benchmark baselines for the pluggable object-store work, documents the results, and adds equal-sample enforcement for base-vs-current measurements. Its latest round also adds an opt-in Linux cache-drop step intended to stabilize peak-RSS measurements, but that step silently lacks permission on the factory VM.

    diff/worktree@synth-10k peak RSS: 13,549,568 bytes → 13,594,624 bytes (0% worse)merge/clean@synth-10k peak RSS: 17,915,904 bytes → 18,190,336 bytes (2% worse)switch/branch@synth-10k peak RSS: 14,753,792 bytes → 14,974,976 bytes (1% worse)
issues fixed5
  • ›grit clone ../src stores un-normalized remote URL with '..' segments (git stores clean absolute path)#2562a20a5666

    Local-path clones now store an absolute origin URL using the caller-supplied environment, matching system Git’s logical-cwd behavior while preserving literal dot components such as ../. Missing local-source errors use that same logical absolute path.

  • ›show/log --stat column bar `|` misaligns when file names contain wide (CJK/emoji) characters#2567e27de440

    Diffstat rendering now measures filename columns with Unicode terminal width and adds explicit spaces, so CJK and emoji paths no longer shift the `|` separator. The change affects only human CLI stat output; JSON, public APIs, repository formats, and protocols are unchanged.

    Randomized debug `grit show --stat` median: 7.7 ms over 300 runs → 7.7 ms over 300 runs (1% better)
  • ›Windows: clean-tree grit status on rust-lang/cargo (3k files) takes 130 ms vs git 47 ms (2.8x)#2566a75b4f92

    This change makes Windows status reuse one FindFirstFileExW directory snapshot for tracked and untracked scans and consumes WIN32_FIND_DATA stat fields instead of opening each tracked path. The round-2 commit repairs all Windows compilation errors in that implementation without changing its design.

  • ›Windows (autocrlf=true): restoring a 1.5 GB file peaks at 3.1 GB RSS (git: 1.5 GB); autocrlf=false streams#2565274e6266

    Large checkout/restore blobs no longer require whole-blob buffering merely because CRLF conversion might apply. Content-dependent autocrlf paths stream through a temporary spool, detect binary/no-op content without a second full allocation, and perform LF-to-CRLF conversion as a streaming second pass only when required; round 3 corrects CRLF handling at input chunk boundaries.

    256 MiB autocrlf restore peak RSS: 572,154 KiB → 50,062 KiB (11× better)256 MiB autocrlf restore wall time: 2,237 ms → 1,041 ms (2.1× better)Peak RSS relative to system Git: 2.1 ratio (system Git = 1.0) → 0.187 ratio (system Git = 1.0) (11× better)Wall time relative to system Git: 4.0 ratio (system Git = 1.0) → 1.9 ratio (system Git = 1.0) (2.1× better)
  • ›grit fsck loads every blob whole: 1 GB file = 1.06 GB peak RSS (git fsck: 10 MB)#2564dfffb5fb

    Large blobs are now verified by fsck without retaining their complete inflated payload: loose blobs stream through the object hasher, and eligible packed whole/delta leaf blobs hash while inflating or applying deltas. The round-3 correction makes the loose corruption regression initialize its repository before writing configuration, so the test fixture is valid.

other changes7
  • ›Main keeper: make CI on main green (clippy, semver-check, rustdoc, public-api)a9489cae

    Refreshes CI-derived API artifacts after the already-landed `repo.merge()` consolidation, points the benchmark-driver rustdoc at the new `Merge::predict` API, and permits the legacy eight-argument integration-test helper under clippy. No production Git behavior, repository format, wire protocol, tests, or removed compatibility components change.

  • ›Main keeper: make CI on main green (public-api, test, semver-check)e1bbc863

    This change repairs CI fallout from the recent 0.6 API consolidation without restoring deprecated public merge functions. It routes integration tests through test-only handle wrappers, updates index-load call sites, and advances the API/semver baselines to the current intentional surface.

  • ›Main keeper: make CI on main green (fmt, public-api, test, semver-check)5e5d1aac

    This test-only patch updates one concurrency test to pass `IndexLoadOptions::default()` after the index-loading API changed. It also adds a `CommitOptions` import to a merge test, although that test still uses the fully qualified type and the import is unused.

  • ›Main keeper: make CI on main green (clippy, public-api)19a38105

    Removes two redundant integer conversions rejected by the pinned clippy toolchain and refreshes the committed public-API snapshot for an already-public worker-budget function. This is a mechanical CI repair with no runtime behavior change.

  • ›Main keeper: make CI on main green (clippy, public-api)09663c9f

    Updates the API-quality baseline for an export already present on main and repairs library-site links broken by the API consolidation. It also updates the site resolver test for the renamed `BranchDeleteMode`; no executable behavior changes.

  • ›Main keeper: make CI on main green (public-api)09663c9f

    There is no integration payload: the reviewed branch and origin/main are the same commit, 09663c9fa. That main commit already raises the public API line ratchet to 32,055 and repairs related rustdoc links/site expectations.

  • ›Follow-up: Before/after benchmark sign-off153850a5

    This branch adds verifiable Linux cache dropping, interleaves system Git, candidate grit, and baseline grit benchmark samples, refreshes the pluggable-store census, and reduces successful checkout-journal allocation work. The evidence harness works, but its refreshed report still does not satisfy the stated RSS gate and one triplet has unequal sample counts.

    status/clean peak RSS: 12.1 MiB → 12.6 MiB (4% worse)merge/clean peak RSS: 16.8 MiB → 17.4 MiB (3% worse)status/clean median wall time: 6.4 ms → 6.4 ms (0% worse)merge/clean median wall time: 49.9 ms → 50.0 ms (0% worse)
$ git log --oneline 90f552da..v0.5.32compare on GitHub ↗
153850a5Interleave baseline grit in daily census for fair peak RSS
6469f917Regenerate pluggable-store after census with verified cache drops
66376526Fail closed when GRIT_BENCH_DROP_CACHES cannot drop page cache
09663c9fFix CI: api-ratchet baseline and library doc rustdoc links
19a38105Fix clippy useless conversions and refresh parallel public API snapshot
f49de4d8Refresh pluggable-store census with drop-cache hygiene and baseline parity test
3a031b9fRefresh pluggable-store after census and sign-off notes
8a7a3cacFix daily baseline-grit parity and checkout journal spool
acfc2f10Record pluggable-store 0.6 headline benchmarks and docs
be1d941cMerge branch 'snapshot-handle-graft'
ba0ef5b9Expose graft, project and filtered_tree on the snapshot checkout handle
9757945eMerge branch 'concurrent-threads'
0488ac51Skip hopeless large deltas and store incompressible payloads
22c16d8fShare worker threads across concurrent grit processes
612d51e3Walk excluded history only to the boundary when writing bundles
a6e86a2cClose out API stragglers on repository handles
d7e44588Record text diff measurements before and after the options struct
dfffb5fbMerge branch 'factory/fsck-stream-large-blobs'
1dc9217dMerge branch 'factory/fix-2565-autocrlf-stream'
e4cc9f5dInit repo before config in loose fsck hash-mismatch test
8df802ddFix fsck index blob presence check and ODB guard test
4c858abdFix fsck peak memory on large blobs
ef041720Fix CRLF split at stream chunk boundary in stream_lf_to_crlf
340b6ab5Buffer LF→CRLF streaming smudge to fix autocrlf restore throughput
f09f10c3Fixes #2565: stream large checkout under autocrlf without double buffering
274e6266Pin the semver baseline to the text diff options API
e819431bRecord text diff measurements before and after the options struct
77856f08Match funcname lines with one regex search and one allocation
e6ad11a1Benchmark text diff counting and rendering in the diff bench
3bfa5beeRefresh the public API snapshot and ratchet after the text diff options
88d7c573Document text diffs and their options in the diff guide
5c903c64Pass combined-diff formatting through options structs
9cacf664Render text diffs through one UnifiedDiffOptions
a75b4f92Merge branch 'factory/win-status-fscache-2566'
adcd79befix: Windows compile for listing cache and align stat probe
38e43bf1perf: share Windows FindFirst listings for status (Fixes #2566)
e27de440Fix --stat pipe alignment for wide path names
5e5d1aacFix IndexFile::load call and merge_df CommitOptions import left off main CI restore
e1bbc863Restore main CI: handle merge tests, index load, api ratchet
a20a5666Remove unused import after cwd prefix cleanup
3d07fb5eUse Environment-only cwd for clone origin URL prefix
0e169f03Fixes #2562: store local clone origin URL like Git absolute_pathdup
d450ed5cAdd commits/status module doctests and handle rustdoc
cb129b65Migrate full-test callers to repo index, status, and commits handles
7ae39c88Fix review blockers: parse fixes, handle callers, create_staged routing
faa697deMove status, staging, and commit APIs to repo handles
c6a2f402fix: semver-compat type aliases and refresh public-api after handle land
e57903e4feat(repo-handles): branches, tags, checkout, notes on handles (#7277)
a9489caeRestore main CI after repo.merge() consolidation