Snapp Technology
Snapp iOS Weekly
Hi folks,
This week has that nice, slightly chaotic feeling where the platform is moving under our feet and everyone is trying to build better footing at the same time. Swift is getting sharper at the runtime edges, WebKit and Xcode are opening doors for agents to actually inspect what our apps render, and the indie side is still doing the hard work: testing on real devices, trimming docs, shipping anyway. The thread, for me, is discipline. New tools are great, but the wins here come from measuring, constraining, and making the machine prove what it claims.
Enjoy!
Articles
Swift
Swift for Wasm September 2026 Updates
Swift on WebAssembly keeps feeling less like a side quest and more like real infrastructure work. WasmKit got a big speed bump, uWASI and JavaScriptKit keep filling in practical runtime gaps, and ElementaryUI is now pushing Swift UI ideas into custom elements. If our codebase ever needs to live outside Apple platforms without becoming a rewrite, this is the kind of steady, boring-in-the-best-way progress we want to see.
withTaskCancellationShield: Swift 6.4 new feature
withTaskCancellationShield is one of those small Swift concurrency additions that immediately answers a bunch of “yes, but what about cleanup?” questions. The useful bit is not magic cancellation prevention; it is hiding cancellation from a block so the work inside can finish cleanly, including child tasks. The catch is the new runtime requirement on iOS 27 and friends, so the back-deployment workaround discussion is the part many of our apps will care about first.
Inside iOS 27’s Reworked Stub Islands
A proper deep dive into the dyld shared cache is not everybody’s idea of a cozy evening, but this one pays off. iOS 27 appears to remove redundant stub sections, rethink trampoline patterns, and scatter objc_msgSend copies through the cache so short branches can still reach. We may never touch this directly in app code, sure, but understanding where Apple squeezes performance out of the system makes the platform feel a little less like a black box.
AI / ML
Getting started with System One models in Swift
System One models are pitched here as “smart if statements”, and honestly, that framing helps. Instead of asking a full LLM to generate text when all we need is a bounded judgement, Jev-style classifiers return typed answers, probabilities, and confidence fast enough to sit in real app flows. For our apps, this is the more interesting AI shape: not chat everywhere, but tiny decision points where fuzzy input meets structured output.
Letting AI See SwiftUI: Xcode Preview MCP in Practice — Pitfalls and Hopes
Xcode Preview MCP sounds like exactly what agentic SwiftUI work needs: let the agent render the UI instead of guessing from code. Then reality shows up. Device selection is shaky, stale screenshots can slip through, and the workaround trail includes deprecated PreviewProvider, fingerprints, and extra metadata so an agent can know it is looking at the right thing. Messy? Definitely. Still probably the future of serious AI-assisted UI work.
Introducing the Safari MCP server for web developers
Safari’s MCP server gives coding agents a browser window they can actually inspect: DOM, console output, screenshots, network requests, page content, interactions, the whole debugging loop. That matters even for those of us living mostly in native apps, because it points to the direction tools are going — agents with eyes, not just autocomplete with confidence. Fewer “try this and tell me what you see” loops is a very easy thing to root for.
How I Built and Shipped a Swift App With Claude Code
Building a Swift app from zero iOS experience with Claude Code sounds like bait, but the story lands because it is stubbornly unglamorous. Specs, plans, CI, UI tests, shell-command hooks, localization checks, real-device testing — all the annoying scaffolding is what kept the agent useful instead of just fast. The best line hiding in here is that the repository has to remember what the agent will forget, which, yes, applies to our human teams too.
Tips
How I reduced my DocC documentation web output size with 95%
Who doesn’t like deleting 855 MB of generated documentation with one build setting? DOCC_SKIP_SYNTHESIZED_MEMBERS=YES stopped DocC from producing pages for inherited SwiftUI protocol members, taking KeyboardKit’s output from roughly 900 MB to 45 MB. If our documentation site is mostly Apple’s view modifiers wearing our package name, this is the kind of boring switch that suddenly makes builds, deploys, and navigation sane again.