fixup! feat: add Git-aware dua clean subcommand

Simplify clean refresh and candidate bookkeeping (change `wtrtlqx`).

- Resolve clean refresh targets directly, preserving whole-candidate Git
  revalidation and the ignore-pattern scope of their original discovery root.
- Share filesystem path construction and candidate-provenance removal with
  traversal code, and reuse ignore-pattern rebasing during discovery and sizing.
- Remove redundant root total updates before cleanup grouping.

Validation: `cargo test --offline clean_`, `cargo test --offline --bin dua`,
`cargo fmt --all --check`, and `git diff --check`.

Diff-stat: 160 insertions, 221 deletions (61 fewer lines).

# d773114 fixup! feat: update deletion progress without blocking navigation

fixup! feat: update deletion progress without blocking navigation

Give `dua clean` a dedicated hub for discovering and grouping deletion candidates.
This follows change `nokwszk` so deletion progress can use the same flat, stable
candidate roots while the user navigates between the hub and the browser.

- Title the hub with its search inputs, group siblings and contained candidates,
  and preserve the selected path as candidates arrive, disappear, or leave a
  singleton. Keep the hub in largest-first order with its own limited controls.
- Open a candidate or group at a detached browser root backed by existing real
  nodes. Reuse ordinary sorting, glob search, Gitignore annotations, navigation,
  marking, and deletion without copying or reparenting candidate trees.
- Keep group membership, group marking, labels, and refresh policy in `CleanHub`.
  Remove group metadata and action-specific clean branches from ordinary entries,
  widgets, and handlers. The explicit hub adds code but reduces browser coupling.
- Preserve whole-candidate Git revalidation and original ignore-pattern scope on
  refresh. Restore the original search directory before replaying a glob, and
  keep navigation inside the scoped results when their viewed entry is deleted.
- Clear browser rows and annotation messages when returning to the hub, so
  deletion cannot consult removed entries after leaving the browser.

Validation:

- Clean journeys cover input titles, grouping, scoped browsing/search, Gitignore
  annotations, refresh provenance, selection, and deletion safety.
- `make check-pre-push`
- `cargo clippy --offline --workspace --all-targets --all-features -- -D warnings`
- Linux and Windows workspace compilation with all targets and features.
- Independent review of scoped navigation, refresh, and deletion. Codex CLI review
  remains unavailable because of expired authentication and a reported usage limit.

# df12220 fixup! fixup! feat: update deletion progress without blocking navigation

fixup! fixup! feat: update deletion progress without blocking navigation

Reduce the UI adaptation introduced in change `mzunukt` while keeping cleanup
grouping separate from ordinary interactive browsing.

- Render borrowed entry rows with presentation flags and an optional label.
  The shared widget consumes an iterator and a selected row position; it no
  longer needs filesystem indices or the application's mark and annotation maps.
- Remove the hub's cloned entry bundles, temporary maps, and fake row indices.
  Use the usual directory marker before appending a group label.
- Share path opening, deletion guards, and opener errors between the hub and
  browser. Keep the distinct partial-group marking rule inside `CleanHub`.
- Simplify hub state access without adding clean-specific branches to ordinary
  browser handlers or changing traversal and deletion scopes.

Validation:

- `cargo test --offline --all-features --bin dua --quiet`
- `cargo clippy --offline --workspace --all-targets --all-features -- -D warnings`
- Regression assertions for fully and partially marked group rendering.
- Independent review of rendering, navigation, group marks, and deletion guards.

Author: codex

README.md

@@ -257,12 +257,12 @@ LANG=zh_CN.UTF-8 dua i   # Simplified Chinese interface
 
 ### Cleanup Mode
 
-`dua clean [DIRECTORY]...` searches for disposable directories and opens the same interactive
-interface. Multiple candidates with the same parent are grouped under that parent, which also
-contains any deeper candidates beneath it. These virtual groups show only cleanup candidates;
-marking a group marks its candidates, leaving the parent and other contents untouched. Single
-candidates without a containing group stay as separate roots. With no paths, it searches the
-current directory.
+`dua clean [DIRECTORY]...` lists disposable directories in a hub titled with the search inputs,
+largest first, without sorting controls or glob search. Siblings are grouped under their parent,
+including any deeper candidates beneath it; other candidates stay separate. Opening a group
+shows its candidates in the normal interactive interface, with sorting and glob search scoped
+to that group. Going back returns to the hub. Marking a group marks only its candidates,
+leaving the parent and other contents untouched. With no paths, it searches the current directory.
 Discovery reads directory entries without file-size metadata, prunes candidates from further
 search, and sizes each candidate completely before listing it. Candidates share the sizing workers
 and appear as each finishes, allowing faster scans to overtake slower ones within the configured

src/clean.rs

@@ -217,9 +217,8 @@ fn is_excluded(path: &Path, root: &Path, cwd: &Path, options: &crate::WalkOption
         .iter()
         .any(|ignored| path.starts_with(ignored))
         || options.ignore_patterns.as_ref().is_some_and(|patterns| {
-            path.strip_prefix(cwd)
-                .or_else(|_| path.strip_prefix(root))
-                .is_ok_and(|relative| patterns.is_excluded(relative, true))
+            crate::pattern_relative_path(path, cwd, root)
+                .is_some_and(|relative| patterns.is_excluded(relative, true))
         })
 }
 

src/common.rs

@@ -410,7 +410,7 @@ pub fn canonicalize_ignore_dirs(ignore_dirs: &[PathBuf]) -> BTreeSet<PathBuf> {
 /// the directory it was given, if it was given exactly one - so that is what patterns should see.
 /// Roots outside via `traversal_root` of the current directory have no such path, and fall back to
 /// being relative to the root they were found under.
-fn pattern_relative_path<'a>(
+pub(crate) fn pattern_relative_path<'a>(
     path: &'a Path,
     cwd: &Path,
     traversal_root: &Path,