Byron/dua-cli v2.45.0

on GitHub

This release packs an incredibly useful new sub-command called dua clean [--depth N],
to automatically and efficiently find typical build products that you may want to delete,
for Rust and Zig and Node for now.

Also, when deleting entries, the TUI does not block anymore, but shows the decreasing number
of bytes as files are deleted, until the marked entry finally disappers. The TUI can still be
used then, albeit in read-only mode.

Oh, and it also got faster on macOS when there are folders with a lot of files, when it will
switch aways from bulk-reads to multi-threaded stat calls instead, which in my case led to
a 2.5x speedup.

New Features

• add Git-aware dua clean and background deletion
  This commit was mostly rubber-stamped, but tested for a week or so before that.
  By now it's fair to accept that dua is in "tool that needs to work, without
  needing to be very familiar with how it works" territory, while I still look
  at the code at least but without the requirement to understand it all.

  Add dua clean [--depth N] [DIRECTORY]... to discover disposable build
  artifacts and caches, size accepted candidates, and offer them in a dedicated
  cleanup hub. Keep the TUI responsive during scanning, deletion, and trashing.

  Discovery and safety

  • Search directory entries and types before collecting size metadata. Recognize
    Node dependencies, Python caches and virtual environments, Cargo build output,
    and Zig caches and output. Require project markers for ambiguous names.
  • In Git repositories, require expendable ignore matches and reject tracked
    contents, conflict stages, gitlinks, and sparse ancestors. Consult HEAD when
    the index is absent and reject candidates when validation fails.
  • Reject nested repositories, case variants of Git markers, unreadable contents,
    excluded descendants, and disallowed filesystem boundaries.
  • Feed discovered roots into dua-core's running walker within the configured
    thread budget. Bound active candidates and publish each independently after
    sizing and validation; commit hardlink/APFS accounting only on acceptance.
  • Search without a depth limit by default; --depth bounds discovery. Support
    traversal filters and --once, with no snapshot import/export or parent
    scanning in clean mode.

  Cleanup hub and browsing

  • Title the hub with its input directories, group siblings and contained
    candidates, and keep rows in largest-first order as scan results arrive.
  • Open groups as scoped views of existing candidate nodes. Reuse ordinary
    sorting, glob search, annotations, navigation, and marking within the group;
    keep group membership and controls local to the hub.
  • Group marks target actual candidates. Complete partial marks and toggle fully
    marked groups off, preserving the real parent and unrelated contents.
  • Refresh recorded candidates without widening their deletion scope. Revalidate
    whole candidates with their original discovery roots, restore browser/search
    positions, and clear stale marks before tree indices are reused.

  Background deletion

  • Run deletion and trash work off the main thread, updating remaining bytes in
    the tree, mark pane, and progress message about once per second.
  • Allow navigation, search, sorting, pane controls, and suspend/resume while
    freezing marks and blocking further filesystem actions until the batch ends.
  • Preserve bounded filesystem concurrency, parallel file removal, and removal
    of child directories before parents. Batch tree updates using scanned sizes
    without double-counting overlapping marks or completed subtrees.
  • Keep failed entries with their remaining sizes and errors. On quit, cancel
    pending work, apply in-flight results, and join workers. On event-loop failure,
    cancel and disconnect events before joining. Trash progress advances when
    each native trash call returns.

• minimize right-hand panels with ]
  A very quick review, but one that showed that the GUI probably shouldn't get more complicated.
  Or else it would be batter with more of a layout system than what it has now - manual layout.

  Behavior

  Add a configurable toggle_right_panes action, defaulting to ], that
  minimizes or restores the shared Help and Marked sidebar. Minimized panes
  display their translated names vertically with a blank column on each side:
  three columns for single-width labels and four for double-width labels.
  The main list uses the remaining terminal width.

  • Keep Help and Marked stacked when both exist, preserving their contents.
  • Return focus to the main list on collapse and skip minimized panes with
    Tab. Restoring with ] keeps focus in the main list; ? restores and
    focuses Help without resetting its scroll position.
  • Keep marks and quit protection intact. Do not dispatch deletion actions
    to a minimized Mark pane, and preserve literal brackets in search input.
  • Retain the minimized preference for the session, including while marking
    more entries or temporarily having no right-hand panes. With no panes,
    ] is a no-op and the sidebar consumes no space.
  • Advertise toggle-collapse = ] on the bottom border of each unfocused,
    expanded Help or Mark pane, using the configured key and translated label.
    Hide the hint during search input, when disabled, or when it cannot fit
    between the border corners.

• add directory_suffix option for directory markers

  Keep the ncdu-style leading slash by default and render a trailing slash when
  directory_suffix = true in the configuration.

• localize the interactive interface

  Translate the remaining interactive panes, status messages, progress text,
  prompts, errors, and notifications for every supported locale. Detect the
  session language once and keep application, command, key, path, unit, Git-Glob,
  and mtime names unchanged, while using compact locale-specific action labels
  where they fit better.

• add German help translation
  Add de locale detection and a complete German translation for the interactive
  help pane, with documentation.

  Regression coverage verifies de_DE.UTF-8 and de selection plus English fallback for de_DE.ISO-8859-1.

• add Simplified Chinese help translation
  Add zh locale detection and a complete Simplified Chinese translation for the interactive help pane, with documentation.

  Regression coverage verifies zh_CN.UTF-8 and zh selection plus English fallback for zh_CN.GB18030.

  Validated with: cargo test interactive::widgets::i18n::tests

• add Korean help translation
  Add Korean locale detection and a complete Korean translation for the interactive help pane. Generalize the existing UTF-8 locale parser so supported language codes share one path, while preserving English fallback for non-UTF-8 locales.\n\nRegression: ko_KR.UTF-8 and ko now select Korean; ko_KR.EUC-KR falls back to English.\n\nValidated with: cargo test interactive::widgets::i18n::tests

• repaint screen with Ctrl-L
  Interactive sessions had no way to recover when terminal contents were corrupted
  because unchanged frames produce no backend writes.

  Add a configurable Ctrl-L repaint action. It clears the backend directly
  to avoid Terminal::clear's cursor-position query racing the input thread,
  invalidates Ratatui's cached frame, and redraws through the normal event path.

• configure flamegraph SVG rendering

  • Add palette, width, minimum-frame-width, title, and inverted-layout flags.
  • Label generated graphs in disk-usage terms with byte counts and paths.
  • Validate rendering arguments and exercise them in parser and end-to-end tests.

• add stacks and flamegraph commands

  Summary

  • Promote folded output previously at dua aggregate --stacks to the first-class dua stacks command
  • Add dua flamegraph, backed by Inferno, with temporary opening by default and --output for file-only output.
  • Keep aggregate --stack as a hidden compatibility path and limit the new commands to relevant traversal flags.

Bug Fixes

• prevent deletion during scans

  Deleting a directory while a scan was active could leave queued traversal
  events referring to removed nodes, aborting dua i with InvalidIndex.

  Reject delete and trash actions while a scan is active, retaining marked
  entries and showing "Traversal already running" so the user can retry
  afterward. Deletion already blocks the event loop, preventing new scans
  until it finishes.

• keep Traditional Chinese locales in English
  Restrict the Simplified Chinese help table to bare zh, zh_CN, zh_SG, and zh_Hans while retaining English fallback for zh_TW, zh_HK, zh_MO, and zh_Hant.

  This addresses the post-commit review finding for 7b5e8e4.

  Validated with: cargo test interactive::widgets::i18n::tests

Performance

• parallelize I/O-bound macOS directory scans
  This was an issue I encountered on APFS, a directory created by rustc with
  more than 1m files in it. In that case, bulk reading is slow (while efficient),
  and it turned out to be better to detect this and switch over to stat-based traversal.

  While looking at the filesystem probing code more closely (and how it's not dependent
  on CPU performance by differntiating wall time from kernel time), I basically
  rubber-stamped all the other code. Too much to look at, too foreign by now.
  But it did look cleaned up, so 👍.

  The reported target-directory scan left one worker waiting inside
  getattrlistbulk at about 15% of a CPU core. Native directory collection also
  held an entire parent directory before publishing entries and child jobs.

  Probe at most two initial bulk buffers before publishing them. Compare
  calling-thread user and kernel CPU time with elapsed time, and select the
  existing parallel stat queue when both refills spend more time waiting than
  executing. Reopen through ordinary enumeration only after that decision,
  discarding the unpublished probe to avoid mixed cursors or duplicate entries.
  Keep bulk reads when the probe is CPU-bound. The initial probe deliberately
  does not adapt to later cache changes.

  Share streaming native traversal between ordering modes, preserve parent
  ordering and APFS metadata, and bound queued metadata jobs by processing
  new batches inline when the queue fills.

  Regression coverage checks scale-independent timing decisions, bounded
  probing and metadata backlogs, streamed child jobs, parent ordering, and
  stat/native parity for clones, resource forks, hard links and symlinks.
  The timing test failed against the provisional fixed latency policy; the
  queue-bound test failed without backpressure.

• preallocate snapshot v2 imports

  Snapshot V2 prefixes the record stream with the total node count and native
  path-name bytes. Interactive imports use them for exact reservations while V1
  remains readable and declared requirements are validated.

  ~/dev benchmark

  dua i --import SNAPSHOT --once, release builds; one frozen V1 traversal converted to V2 for identical contents.

  Metric	V1	V2	Change
  Median peak RSS (5 runs)	369,868,800 B	230,473,728 B	-37.7%
  Mean wall time (20 runs)	430.5 ± 1.8 ms	438.1 ± 2.8 ms	+1.8%
  Compressed snapshot size	38,215,358 B	38,215,404 B	+46 B

Performance (BREAKING)

• avoid unnecessary metadata during deletion

  Deletion needs entry types, but every walk collected full sizes, timestamps,
  allocation information, and file identities. Add Options::skip_metadata and
  use directory-entry types without extra metadata lookups when available.
  macOS and Windows use standard enumeration for this mode; Windows roots
  query only attribute/tag information. Keep the native metadata readers for
  ordinary scans and preserve Windows verbatim-path handling.

  Drop per-entry deletion byte accounting. Successful removals use the existing
  scanned total; partial failures retain entry/error counts and display unknown
  bytes instead of collecting metadata solely for that notification.

  Local macOS/APFS release medians showed 0-11.5% lower deletion time, with
  wide-directory traversal up to 40% shorter; gains vary with tree shape.

Commit Statistics

• 26 commits contributed to the release over the course of 13 calendar days.
• 13 days passed between releases.
• 15 commits were understood as conventional.
• 1 unique issue was worked on: #392

Commit Details