[Trap101]

What

An optional Spaces-style slide when switching virtual workspaces. The outgoing workspace slides off one edge while the incoming one slides in from the other. The direction follows workspace order, and next/prev keep their natural direction across wrap-around. It's off by default and independent of animate.

workspace_transition = "none"            # "none" | "slide"
workspace_transition_duration = 0.25     # seconds
workspace_transition_axis = "horizontal" # "horizontal" (next enters from the right) | "vertical" (next enters from below)

We're open to changing this. If you'd prefer a different shape, for example a nested [settings.workspace_transition] table, folding it into the existing animation_* keys, or different names and values, we're happy to discuss and rework it.

How it works

• Built on existing pieces. It uses the presenter and transition engine from perf: smoother animations #551, so there's no new thread or timer. apply_layout routes the switch arrange to a new AnimationManager::workspace_transition_layout when a direction is pending for that space.
• Which windows move. Windows are classified by geometry, not workspace membership. A window is incoming if its target is visible and it's currently parked, and outgoing if the reverse. Everything else, including the dragged window, goes through the existing instant path.
• Ending at the normal park. Outgoing windows slide offscreen, keeping 1px revealed, and settle at the usual hidden-corner park on the final frame. The transaction target is the normal park, so the rest of the system sees the normal parked state, and on multi-display setups windows always come to rest at their regular park.
• Detecting parked windows. A new HiddenWindowPlacement::is_parked treats a window as parked if at most a sliver shows on either axis. macOS pulls parked windows back up by their title bar (a park requested at y=955 lands around y=924), so an exact match fails.
• Fallbacks. The switch stays instant during a drag and in low-power mode.
• Not cancelled by follow-up arranges. instant_layout_inner now only cancels presentations whose pending target actually changed, so focus and raise arranges right after a switch don't cut the slide short.

Rapid switching

Toggling workspaces quickly (⌥1 ⌥2 ⌥1 …) was expensive, so two changes affect the shared engine:

1. A clock per window in ActiveAnimation.
   • Previously every Replace restarted all windows on a fresh full duration, including windows that weren't retargeted.
   • Now a retargeted window continues from its presented frame, takes a share of the duration matching the distance it has left, and eases out instead of starting from rest.
   • Windows the replacement doesn't touch finish on their original schedule.
   • Each window ends its lease as soon as it lands, and finished windows are no longer re-sent every tick.
   • This also affects animate = true layouts, since they share replace_with.
2. No repair for parked windows.
   • A settled park frame is tagged FrameSource::Park.
   • The app actor skips its end-of-motion repair for it. macOS always clamps a park, so the repair could never succeed.
   • Over 10 slides this removed 15 of 15 pointless set_frame writes, one per outgoing window. Other AX traffic was unchanged.

Testing

• cargo +nightly fmt --all --check, cargo check --locked, and the full test suite pass.
• New tests:
  • config defaults, the template, and validation;
  • direction, including wrap-around and no-op switches;
  • incoming and outgoing frames on both axes;
  • a slide from an empty workspace;
  • instant re-arranges not cancelling an in-flight slide;
  • retarget duration scaling;
  • carried windows keeping their own clock;
  • per-window lease end;
  • a landed window restarting from the caller's frame rather than its park;
  • park frames being tagged Park.
• Manual: used day to day on a single 1470×956 display with the traditional layout, including rapid toggling. I haven't tested a multi-monitor setup yet.

🤖 Generated with Claude Code

https://claude.ai/code/session_018JyfUbLLYEiyoxyQ2vjRsy

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.