New Features

• reuse workers when restarting a walk

  Retain a completed single-root walk pool and add Walk::restart() so callers
  that rescan a changing directory avoid rebuilding worker threads.

  Declare Rust 1.88 support for let-chain and integer-cast syntax.

Commit Statistics

• 2 commits contributed to the release.
• 3 days passed between releases.
• 1 commit was understood as conventional.
• 0 issues like '(#ID)' were seen in commit messages

Commit Details

Thanks to our contributors, there are not one, but two (a joke, a human is writing this with fleshy fingers :D) headline features:

1. dua aggregate --depth N shows an indented tree-like view, instead of a flat list
2. dua aggregate --stack | inferno-flamegraph > fire.svg outputs a flame-graph compatible stack listing, for pretty visualisations of the directory. It also works with --depth N.

On the side, on macOS, the bytes output of dua now matches du byte-perfectly, and for the first time it's possible to pipe dua
output into a file without terminal escape sequences (why did this take me so long again?).

New Features

• Add dua aggregate --depth N for indented tree output

Bug Fixes

• don't output colors when stdout isn't a terminal.
  This facilitates piping into a file.
• --stats printed the u128 sentinel as the smallest file size
  The sentinel reset was guarded by entries_traversed == 0, but an entry
  whose metadata cannot be read still counts as traversed while never
  contributing a size. Reset when the sentinel is still in place instead.
• Apply --ignore-dirs to prepared aggregate roots
• Apply --ignore-dirs to paths dua expanded into roots itself

Commit Statistics

• 28 commits contributed to the release over the course of 9 calendar days.
• 10 days passed between releases.
• 5 commits were understood as conventional.
• 0 issues like '(#ID)' were seen in commit messages

Commit Details

New Features

• Match macOS du allocation accounting
  Mirror Apple FTS complete common/file attribute request and packed-invalid records so macOS allocation totals match /usr/bin/du. Reject unknown vnode types, round allocation bytes to 512-byte stat blocks, and fall back to symlink metadata for directories, firmlinks, and incomplete records.

  Layer optional APFS clone and data-fork attributes onto the same packed layout so the upstream clone-deduplication path remains intact.

Bug Fixes

• don't output colors when stdout isn't a terminal.
  This facilitates piping into a file.
• a relative root ending in .. is mis-resolved

Commit Statistics

• 12 commits contributed to the release over the course of 9 calendar days.
• 10 days passed between releases.
• 3 commits were understood as conventional.
• 0 issues like '(#ID)' were seen in commit messages

Commit Details

*the same as 2.42, and:

Starting directories with a large amount of files, like 50k, now see a 5x speedup on macOS and Windows as bulk-reading is also done there. Note also that this is still a small absolute difference, 100ms vs 500ms, but a good demonstration of how much large trees with a lot of such directories will benefit by this, as these small absolute improvements accumulate.

Commit Statistics

• 3 commits contributed to the release.
• 1 day passed between releases.
• 0 commits were understood as conventional.
• 0 issues like '(#ID)' were seen in commit messages

Commit Details

Support providing walk roots with their metadata already set due to bulk-reading.

Commit Statistics

• 4 commits contributed to the release.
• 1 day passed between releases.
• 0 commits were understood as conventional.
• 0 issues like '(#ID)' were seen in commit messages

Commit Details

The headline or this release is ~30% better traversal performance on macOS due to the usage of bulk-metadata APIs on supported filesytems.

Bug Fixes

• sanitize control characters in marked path output
  dua-cli's interactive TUI is built on ratatui, which protects the
  paths it renders on screen. But marking a file for deletion and then
  quitting prints that file's path directly to the terminal after the
  TUI has already released terminal control, bypassing ratatui's
  protective rendering entirely. A scanned file's name has no character
  restrictions, so a crafted file name can inject terminal escape
  sequences into the printed path.

  Add sanitize_for_display (src/main.rs), replacing control characters
  with the Unicode replacement character, applied at the sole print site
  for marked paths.

Commit Statistics

• 8 commits contributed to the release over the course of 9 calendar days.
• 10 days passed between releases.
• 1 commit was understood as conventional.
• 0 issues like '(#ID)' were seen in commit messages

Commit Details

macOS specificy bulk-readtree implementation for 30% performance boost on supported filesystems.

Commit Statistics

• 5 commits contributed to the release over the course of 9 calendar days.
• 9 days passed between releases.
• 0 commits were understood as conventional.
• 0 issues like '(#ID)' were seen in commit messages

Commit Details

A one-line change that makes all the difference: Windows TUI performance isn't sluggish anymore as the drawing is now buffered!

Bug Fixes

• buffer terminal frame output

Commit Statistics

• 5 commits contributed to the release.
• 1 day passed between releases.
• 1 commit was understood as conventional.
• 0 issues like '(#ID)' were seen in commit messages

Commit Details

There are two major features: 30x and more performance on Windows, and --ignore-from <file> support.
This makes this release the best one yet, and I do hope that I can last a week or more until the next one.

New Features

• add --ignore-from to exclude paths with gitignore-style patterns
  Reads gitignore-syntax patterns from one or more files and leaves everything
  they match out of the report, in both aggregate and interactive mode. This is
  the equivalent of rsync's --exclude-from and restic's --exclude-file, so the
  same pattern file can answer "how much of this would actually be backed up?".

  Matching is powered by gix-ignore, which is already a dependency, so negation,
  anchoring, ** and directory-only patterns all behave exactly like Git.
  Excluded directories are pruned from the walk rather than only hidden, and
  excluded top-level paths are dropped before the walk so they are absent from
  the report instead of appearing as empty.

Bug Fixes

• propagate background root device errors

  Background traversal replaced failed root-device lookups with device ID
  zero. Readable root metadata could then be rejected as cross-device while the
  traversal incorrectly retained a successful error count.

  Skip roots whose lookup fails and carry their per-root errors in the traversal
  completion event, restoring the previous statistics and exit status. Keep
  interleaved overlapping and duplicate roots isolated by including each root path
  Arc allocation in private directory-map keys while preserving public traversal
  event types.

• propagate aggregate root device errors

  A failed root-device lookup was replaced with device ID zero after the parallel
  traversal rewrite. For a dangling symlink, symlink metadata remained readable
  but failed the device check, hiding the initialization error and producing a
  successful exit status.

  Count the lookup failure on its root, mark that root complete for ordered
  output, and omit it from traversal. Flush completed roots after traversal
  because failed roots emit no completion event.

  Keep all roots in the shared
  parallel traversal pool; when hard links are deduplicated, per-root attribution
  is scheduling-dependent while the total remains correct, in the name of performance.

Commit Statistics

• 14 commits contributed to the release.
• 2 days passed between releases.
• 3 commits were understood as conventional.
• 1 unique issue was worked on: #302

Commit Details

Even more performance, up to 15%. That's it now, trying to hit 1 week without a release ;).

Bug Fixes

• avoid aggregate progress flicker Track whether aggregate progress has actually been rendered before clearing the
  terminal line. Keep sorted multi-root progress visible until final output, while
  clearing before immediate unsorted results, and cover fast roots that previously
  emitted repeated invisible erase sequences.

Performance

• parallelize filesystem traversal

  Split completion-order directory entries into small stealable metadata jobs
  and use idle-aware relay waking with LIFO local queues. Submit all top-level
  roots to the same worker pool, carrying a root index through descendant jobs and
  result batches so aggregate and interactive traversal retain per-root accounting
  without serial barriers.

  Also, track pending work per root and emit completion events after all of a
  root's batches are delivered. Flush the longest completed input-order prefix so
  --no-sort prints results immediately without serializing the shared worker pool.

Commit Statistics

• 7 commits contributed to the release over the course of 1 calendar day.
• 1 day passed between releases.
• 2 commits were understood as conventional.
• 0 issues like '(#ID)' were seen in commit messages

Commit Details