Day

Status: resolved design discussion. See the linked app reference for current API guidance.

Focus design decisions

6. Prior art: what Day adopted and rejected

  • SwiftUI (@FocusState, .focused(_:equals:)). Adopted: the Bool + Optional-of-Hashable binding shape, nil clears focus and dismisses the keyboard, moved focus writes back on loss. Rejected: the unconstructible binding (state can’t live outside the view); Day signals have no such restriction.
  • Flutter (FocusNode/FocusScope). Rejected as an API (imperative node lifecycle inside a declarative tree); adopted as semantics: “focus changes apply after the build phase” is rule 3, and the scope-restore behavior informs the reserved focus-scope design.
  • floem (the nearest cousin): its request_focus(when) showed that signal-driven focus writes work; its event-only reads are the asymmetry rule 2 closes.
  • iced / egui / Slint / GPUI: converge on last-write-wins, next-frame application, and focus-nowhere being representable, all reflected in the rules.

7. Resolved design questions

  1. Group-signal loss semantics — solved in the pump: a FocusChanged(false) scans the queue for a paired FocusChanged(true) from another node and dispatches the gain first, so group signals never blip through None (rule 6).
  2. Resign target on Android/XAML — both resign to a focusable root: Android to the activity’s focusable-in-touch-mode content root, XAML to a hidden focus-sink control.
  3. focused_eq naming — one .focused() name via two marker-trait impls; the tuple form (signal, key) replaces a second method.
  4. Headless CI focus — programmatic grab_focus/setFocus is window-local on every backend (Qt after the app-local activation above), so scripted runs don’t depend on the window manager granting real OS focus.