iOS 27 API support: adaptive layout, NowPlaying, Foundation Models #5536

Summary

iOS 27 / iPadOS 27 / macOS 27 betas landed (beta 1: 2026-06-08); GM expected ~Sept 2026. This issue tracks adopting the iOS 27 APIs worth supporting in Perry's native UI/runtime layer. Not urgent — work this as iOS 27 lands. Nothing here can ship to the App Store until Xcode 27 GM (Apple rejects beta-SDK submissions), so treat it all as build-and-experiment for now.

Companion concern (SDK/toolchain plumbing, separate from API adoption): native on-Mac builds auto-adapt to the installed SDK via xcrun, but the Linux cross-compile path pins the SDK version (26.0.0) in crates/perry/src/commands/compile/link/platform_cmd.rs (iOS/tvOS/macOS -platform_version args) and the App Store GM DT* markers are hardcoded in crates/perry/src/commands/compile/bundle_apple.rs (gm_dt_defaults, DEFAULT_DT_XCODE). Those are a GM-day bump, not a beta-day one — bumping the DT* markers early would break App Store submissions.

Fit filter

Perry talks to Apple frameworks via a handle-based, imperative FFI (hand-written extern "C" + class!()/msg_send!, frameworks linked at link time — same pattern as the existing AVFoundation/CoreLocation/MapKit integrations). So:

• Imperative, handle-shaped APIs bridge cleanly → good candidates.
• Swift-macro/protocol-shaped APIs (@Generable, AppIntent conformances, entity schemas) are a poor fit → high effort, low payoff, defer.

Proposed scope (ranked)

1. Adaptive layout APIs (SwiftUI/UIKit) — do first

Fold/hinge-state detection, side-by-side, 4:3 aspect, split-view multitasking. This is plain UIKit layout — the area perry-ui-ios already owns (splitview, tabbar, stack widgets). Lowest risk, no new framework, fits the existing widget model. Apps break visually on foldables/multitasking without it.

2. NowPlaying framework — natural extension

Wires playback to Lock Screen / Control Center / Dynamic Island / CarPlay. Perry already has media_playback.rs (AVPlayer) + audio_playback.rs; this is the missing companion. Imperative/handle-shaped (MPNowPlayingInfoCenter-style), bridges like the existing audio FFI. Medium effort, high user-visible value.

3. Foundation Models — strategically the big one, scope carefully

On-device LLM access; in iOS 27 it also fronts cloud providers (Claude/Gemini) via a LanguageModel protocol. Compelling for Perry's audience (on-device AI without leaving TypeScript). Ship the simple async path (session → respond(prompt) → string); skip guided/structured generation (@Generable macros/schemas) initially — that part is deeply Swift-shaped.

Lower priority / poor fit

• App Intents / Siri — now mandatory for Siri (SiriKit deprecating). Strategically real once a production Perry app needs it, but the most macro/protocol-heavy API in the release → worst fit for Perry's FFI. Defer.
• Core AI (load/run your own models) — lower-level than Foundation Models; most TS app devs want the latter. Defer.
• Music Understanding (on-device 6-dimension audio analysis) — niche, audio-app-only. Skip unless requested.
• Game Porting Toolkit 4 — an agentic Metal porting tool, not a runtime API to bind. Out of scope.

Caveats

1. New iOS 27 APIs need either updated objc2-* binding crates (currently pinned 0.3/0.6) or hand-written FFI. Hand-written is the established Perry pattern, so not a blocker — just the work.
2. Beta-SDK = no App Store. Prototype against beta devices now; ship after Xcode 27 GM.

References

• iOS & iPadOS 27 Beta Release Notes — Apple
• What's New - iOS — Apple Developer
• Apple Newsroom — next-gen Apple Intelligence, Siri AI, and more
• Apple Releases First iOS 27 Beta 1 — MacRumors