sweep/2026-10-04: 50 fixes, 15 refactors (36→43 tests)

[Chartres]

Sweep 2026-10-04 — mac-dir-stat

5 cycles · 100 commits · tests: 36 → 43

Findings fixed (50 total: 9H · 28M · 13L)

#	File	Symptom	Sev
1	app.rs	perform_delete no is_alive check; trashes wrong path if node died while dialog was open	H
2	scanner/walk.rs	APFS firmlinks (/Users /Applications etc.) treated as volume roots; fake Free Space injected	H
3	scanner/walk.rs	Dir not in dir_map after metadata fail; children get wrong full_path → Trash hits wrong file	H
4	ui/context_menu.rs	"Move to Trash" offered for scan root and synthetic nodes	H
5	app.rs	Cmd+Backspace no scan-root guard	H
6	scanner/tree.rs	remove_node no dead-node guard; double-call corrupts ancestor sizes	M
7	app.rs	poll_partial_refresh keeps stale ConfirmTrash / ConfirmBatchTrash ids after graft	M
8	app.rs	Cmd+Backspace / Enter not gated by wants_keyboard_input; search-clear opens trash dialog	M
9	scanner/walk.rs	TCC skip list applied even with Full Disk Access granted	M
10	app.rs	Batch-trash failures silently go to stderr only	M
11	app.rs	view_root / zoom_stack / selection not sanitized after delete; dead view_root	M
12	scanner/mod.rs	Panic in scan thread drops sender; UI stuck at "Scanning…" forever	M
13	scanner/tree.rs	remove_node on dead node — now idempotent; test added	M
14	ui/treemap_view.rs	Selecting "(none)" file type dimmed extensionless files (extension None ≠ "")	M
15	app.rs	poll_partial_refresh grafts without checking target alive; sets selected_node to dead id	M
16	scanner/walk.rs	/System/Volumes in skip_paths blocked the scan-root exception → empty tree on Data volume	M
17	app.rs	ConfirmBatchTrash not invalidated after graft	M
18	scanner/walk.rs	Incomplete firmlink list (/usr/local etc.); partial refresh on them doubled disk total	M
19	cleanup.rs	find_candidates could classify the scan root; "Trash All" would trash the entire scan root	M
20	app.rs	zoom_stack fallback after cleanup batch-trash pushed dead root node	M
21	app.rs	Scan / partial-refresh errors went only to eprintln; status_message never set	M
22	scanner/tree.rs	Synthetic node sentinel __ collided with real files like foo.__bak	M
23	app.rs	Cleanup candidate size stale after per-item deletes; batch total overstated	M
24	platform/trash.rs	empty_trash left zombie; Finder failure silent	M
25	app.rs	perform_batch_delete never reset view_root/zoom_stack; dead view_root after cleanup trash	H
26	app.rs	Cmd+Backspace allowed ConfirmTrash on synthetic nodes (context menu was guarded, keyboard was not)	L
27	ui/dir_tree.rs + ui/context_menu.rs	Zoom stack push without duplicate guard (only treemap_view had the fix)	L
28	app.rs	perform_delete kept dead ancestors in zoom_stack	L
29	platform/finder.rs + platform/fda.rs	Spawned open/osascript children never waited → zombie per click	L
30	treemap/squarify.rs	partial_cmp().unwrap() panics on NaN size	L
31	cleanup.rs tests	Fixed temp-dir names → collision when two test processes run concurrently	L
32	flywheel.rs tests	set_var in test mutated env across parallel threads	L
33	flywheel.rs	Duplicate state_dir() (already pub(crate) in state.rs after cycle 1 dedup)	L
34	scanner/walk.rs	/home and /net listed as APFS firmlinks (they are autofs; lookups can stall)	L
35	scanner/walk.rs	Redundant should_skip inside loop body (same check already ran before dir_map insert)	L
36	app.rs	status bar mixed tree file_count (includes synthetic nodes) with stale scan_progress.dirs	L
37	app.rs + context_menu.rs	Synthetic nodes reached Open / Show in Finder / Get Info / Enter-reveal	L
38	scanner/walk.rs	Partial refresh on non-root mount point injected synthetic nodes into tree mid-branch	L
39	app.rs	perform_batch_delete cleared selection only for exact id match; dead child nodes stayed selected	L
40	scanner/walk.rs	find_closest_ancestor re-parented orphans; full_path() returned wrong path → Trash hit wrong file	M
41	scanner/tree.rs	StringArena as u32 cast silently wraps at 4 GiB → corrupt full_path; now panics loudly	L
...	ui/treemap_view.rs, ui/search.rs, flywheel.rs, state.rs, ui/cleanup_window.rs, ui/widgets.rs	various L fixes	L

Contested / skipped (architecture-level)

• Scan thread cancellation (threading redesign)
• Hard link / sparse file / APFS clone size correctness (platform complexity)
• find_candidates on UI thread (background task infra)
• graft_under memory reclaim
• Per-frame dir_tree clone / squarify stack depth
• full_path for entries found via find_closest_ancestor when parent is missing (follow-up: now orphans are dropped instead of re-parented — safer)

Tests

36 → 43 (+7): added tests for remove_node idempotency, graft_under, clear_descendants, recompute_sizes_upward, find_candidates, scan-root orphan guard, and test isolation fixes (temp dirs, env races).

Refactors (15)

• Reuse platform::dialogs::pick_folder in toolbar
• Deduplicate state_dir() in fda.rs and flywheel.rs
• ext_list.rs: iter/enumerate over index loop
• treemap/color.rs: lerp closure
• treemap/mod.rs: dead gap > 0 guard removed
• toolbar.rs: chained display call
• ui/cleanup_window.rs: single pass for batch selection
• ui/dir_tree.rs: frame built once
• platform/finder.rs: spawn_reaped helper (was already fixed inline, now unified)
• ui/widgets.rs: reuse fg for label color
• ui/help_window.rs: reuse tip() helper
• flywheel.rs: dropped unused PathBuf import (only compiler warning)
• ui/treemap_view.rs: 3 simplifications (redundant guard, find loop, is_some_and)

🤖 Generated with Claude Code