DragAndDropResolver — Specs

Summary

DragAndDropResolver is the pure decision layer for drag-and-drop over the switcher: what a drag-over of the tiles should do (dragOver), when the auto-select timer (re)arms (movedBeyondResetRadius), whether the global mouse tap must yield the drop's mouse-up (passesThroughMouseUp), and whether a drop is valid (canDrop). It holds no state and touches no AppKit drag session, timer, or event tap — TilesDocumentView (the NSDraggingDestination) and CursorEvents (the global tap) turn its decisions into real NSDragOperations, timers, and tap pass-through (same pattern as AxEventRouting).

Behavior & edge cases

• Two owners, one boundary. AppKit's NSDraggingDestination owns the drag (hover highlight, the .link operation, the auto-select timer); the global mouse tap owns ordinary clicks. They conflict only on the one event that ends a drag — the leftMouseUp. A file drag over the switcher can only have started in another app BEFORE the switcher showed (the tap swallows any leftMouseDown outside the panel, and tiles aren't drag sources), so the tap never saw that drag's leftMouseDown. passesThroughMouseUp keys off exactly that: yield any up whose down the tap didn't see, so AppKit / the source app concludes the drop instead of the tap swallowing it and leaving the file glued to the cursor. This holds wherever the release lands — on a tile, on the padding around the tiles, or outside the panel — because none of those are a down the tap saw. A normal click's down IS seen, so clicks route normally. This is the #5350-regression fix.
• No grab on appear + deadzone. A drag already in flight when the switcher shows must not select a window on the first stray pixel. dragOver reports .inDeadzone (still a valid .link drop, but no selection and no timer) until the pointer clears the same movement deadzone mouse hover uses.
• The inter-tile gap targets like hover. Targeting reuses hover's findTarget, which expands each tile by 1px so the 1px gap between tiles still resolves to a tile. The kernel only sees hasTarget; it never returns .noTarget while the cursor is over the grid.
• The auto-select timer always runs for a drag. Hover gates the 2s auto-select timer on a preference; dragging is a stronger intent, so it always arms the timer when past the deadzone. The timer (re)arms on a target change or once the pointer leaves the reset radius (movedBeyondResetRadius, inclusive at the radius); a sub-radius jitter within the same tile lets the running timer fire. A nil anchor (timer not armed yet) counts as "re-arm".
• A drop opens files with the tile's app, so it needs a target tile, that tile's window, the app's bundle URL, and at least one URL; a non-URL drag (text/image) or any missing piece is rejected (the drag snaps back).

Test scenarios

Mirrors DragAndDropResolverTests.swift 1:1.

A. Drag-over operation (deadzone, targeting, always-on timer)

• testNoTargetReportsNoDrop — off the grid (no tile) → .noTarget.
• testTargetInDeadzoneLinksWithoutSelecting — tile present but deadzone not cleared → .inDeadzone (no grab on appear).
• testTargetChangeArmsTimer — moving onto a different tile → .track(restartTimer: true), regardless of distance.
• testSameTargetWithinRadiusKeepsTimerRunning — same tile, sub-radius jitter → .track(restartTimer: false).
• testSameTargetBeyondRadiusRearmsTimer — same tile, left the radius → .track(restartTimer: true).

B. Auto-select timer reset radius (the 5px rule)

• testNoAnchorAlwaysRearms — nil anchor → re-arm.
• testWithinResetRadiusDoesNotRearm — moved 4px (radius 5) → no re-arm.
• testAtResetRadiusRearms — moved exactly 5px, and the 3-4-5 diagonal → re-arm (inclusive boundary).
• testBeyondResetRadiusRearms — moved 10px → re-arm.

C. Mouse-up pass-through (the regression fix)

• testPassesThroughMouseUpForUnseenDown — a drag's down was never seen → yield the up (don't swallow the drop).
• testSwallowsMouseUpForSeenDown — a click's down was seen → route normally (not yielded).

D. Drop validity

• testDropNeedsTargetWindowBundleAndUrls — target + window + bundle URL + ≥1 URL → drop.
• testNoTargetRejectsDrop — no target → reject.
• testNoWindowRejectsDrop — target without a window → reject.
• testNoBundleUrlRejectsDrop — app without a bundle URL → reject.
• testNoUrlsRejectsDrop — non-URL drag / empty pasteboard (0 URLs) → reject.

E. Use-case integration (the discussed scenarios as decision sequences)

• testUseCaseDragPresentWhenSwitcherAppears — drag present on show → .inDeadzone; after clearing the deadzone → .track.
• testUseCaseDropOnTileConcludes — the regression: the tap yields the up (down unseen), and a valid target drops.
• testUseCaseReleaseOnPaddingEndsDragWithoutOpening — release on the padding / outside the panel: the up is yielded so the drag ends, but no target means nothing opens.
• testUseCaseBetweenTilesStillTargets — the 1px gap resolves to a tile upstream, so the kernel keeps tracking (never .noTarget over the grid).
• testUseCaseAutoSelectTimerSurvivesJitterButRearmsOnMove — jitter keeps the timer; a real move re-arms it.