Fatbobman's Swift Weekly #148
Apple Intelligence Has Passed Chinese Government Review and Is Coming to China
Apple Intelligence Has Passed Chinese Government Review and Is Coming to China
At WWDC 26, Apple unveiled a new Apple Intelligence architecture. Gemini was introduced on the cloud side, while Apple also confirmed that its third-generation Apple Foundation Models were built in collaboration with Google. For a while, this made me think that bringing Apple Intelligence to China might actually become even more difficult.
But things moved faster than expected. In July, Apple Intelligence appeared on the latest list of generative AI services registered with China’s Cyberspace Administration of China (CAC). Then, just two days ago, Apple confirmed that Alibaba’s Qwen would be involved in the Chinese version of Apple Intelligence, and the integration briefly appeared in Apple’s official documentation. Although the document was temporarily taken down the day after going live on August 8, its detailed setup instructions and privacy descriptions had already revealed a fairly complete picture of how the integration would work. This suggests that the official launch of Apple Intelligence in China is now entering its final countdown.
In addition to Alibaba’s Qwen, the registration information also revealed that Baidu will participate deeply as a partner in the AI search and visual search capabilities of the Chinese version of Apple Intelligence. In other words, Apple can not only adjust the underlying models and service providers by region, but also rely on more than one partner within the same market.
As countries and regions continue to strengthen their requirements around privacy, data, and AI services, regional differences are becoming almost unavoidable. China is becoming Apple’s first major example of a new approach: keeping the overall Apple Intelligence product experience and developer interfaces relatively consistent, while selecting different underlying models based on local regulations and ecosystems. This model is likely to appear in more markets over time and could become an important part of Apple Intelligence’s global expansion.
For developers using Apple Intelligence APIs, this is also good news. If Apple can continue to maintain consistency at the API layer, developers should theoretically not need to maintain completely different implementations for different markets. It would also allow apps that rely heavily on Apple Intelligence capabilities to reach a much broader audience.
Gemini and Qwen may not be the two most prominent names in the Agent scenarios I use day to day, but both perform quite well in areas such as world knowledge and language understanding. Given that Apple Intelligence is currently more focused on system-level assistance, and that these partners also bring mature model infrastructure and extensive cloud service experience, Apple’s choices seem quite reasonable.
Apple Intelligence has become one of Apple’s most important entry points into the AI era. Its current capabilities are still somewhat short of my personal expectations, but starting in 2026, it is at least beginning to reach substantially more users around the world. Rather than focusing only on how many tasks it can complete today, I am more interested in seeing how Apple will use WWDC 2027 to bring AI more deeply into its hardware, operating systems, and services ecosystem.
Previous Issue｜Newsletter Archive
📢 Sponsor Fatbobman’s Swift Weekly
Promote your product to Swift & iOS developers across:
- Blog: 50,000+ monthly visitors
- Newsletter: 4,000+ subscribers, 53% open rate
Perfect for developer tools, courses, and services.
Enjoyed this issue? Buy me a coffee ☕️
Original
Controlling Orphans in SwiftUI Text: Uncovering the Undocumented avoidsOrphans
SwiftUI’s Text has a default behavior: when the last line of a paragraph contains only a single orphaned word or character, it automatically moves an entire word group from the previous line down to avoid leaving it alone. While this behavior is often helpful, it can also leave excessive whitespace on the preceding line, especially in narrow layouts or mixed Chinese–English text, disrupting the visual balance of the paragraph. UIKit allows developers to disable this behavior through NSParagraphStyle, but SwiftUI exposes no public API for doing so. This article uncovers an undocumented SwiftUI API that has existed for quite some time—avoidsOrphans—allowing developers to regain control over this behavior.
Recent Recommendations
An Even Closer Look at Protocols and Global Actors
As Swift’s concurrency model continues to evolve, some engineering decisions that once felt obvious are beginning to change alongside the language itself. In this article, Matt Massicotte goes beyond the familiar question of whether @MainActor belongs on a protocol or individual methods, using it as a starting point to explore API design in the era of Swift Concurrency. Rather than promoting a so-called “best practice,” Matt emphasizes a broader principle: isolation is a constraint, and constraints should arise from real requirements—not from habit.
Weekly Tuist News #1
As AI agents become increasingly integrated into the development workflow, engineering tools such as Tuist are becoming more important—even for independent developers.
Longtime Tuist community contributor Lee YoungJun has launched a new series, Weekly Tuist News, dedicated to tracking the Tuist ecosystem. Rather than simply summarizing release notes, the series combines community updates with the author’s hands-on experience. In its debut issue, he covers SwifterPM becoming the default dependency restoration method, new controls for target cache invalidation, and Tuist’s auth.md support for coding agents, along with practical observations from real-world usage.
Preview Multiple SwiftUI View States with #Preview(arguments:)
Most SwiftUI components have more than one meaningful state: empty, loading, error, permission-restricted, and so on. During development, we often end up maintaining multiple nearly identical #Preview declarations or placing every state into a single VStack. Xcode 27’s new #Preview(arguments:) API is designed specifically for this scenario, generating multiple independent previews from a collection of state data while allowing each one to be inspected individually in the Canvas.
In the article, Artem Novichkov also shares his approach to organizing previews: arguments should describe business states, not environments. Even as the API evolves, this separation between application state and runtime environment remains a valuable design principle.
Concentric Buttons with OS 27’s SwiftUI APIs
WWDC26 introduced GeometryProxy.concentricCornerRadii, allowing SwiftUI developers to access a container’s concentric corner radius information. However, ButtonBorderShape remains a closed type, making it impossible to create truly custom button border shapes directly. Alexander Weiß demonstrates how to leverage this new API together with onGeometryChange, a custom ViewModifier, and method overloading to build a concentric button solution that feels almost indistinguishable from a native SwiftUI API.
For new APIs such as
GeometryProxy.concentricCornerRadii, it usually takes only a few seconds to understand what they do. The real challenge comes when using them in production, where implementation details and framework limitations quickly surface. Rather than serving as an API introduction, this article explores those practical considerations and the scenarios where the API can be effectively applied.
Swift, AI, and the future of software development
AI has become an indispensable part of many developers’ workflows, but beyond improving productivity, it is also quietly changing the relationship between developers and their code. In this article, Tibor Bödecs candidly reflects on his own experience over the past year: embracing AI for its efficiency while missing the satisfaction of writing code and deeply understanding problems; believing that AI will continue to reshape software development while hoping it remains a tool that augments human thinking rather than replacing it.
Over the past months, I’ve found myself struggling with many of the same questions as Tibor Bödecs. While AI has taken away some of the joy and created a subtle sense of distance between myself, the code, and the products I build, I’ve also been working relentlessly to gain greater control over it—to make it an extension of my own thinking, rather than allowing myself to become an extension of it.
Tools
Gitwyn: A Git Client Designed Around the Xcode Workflow
Although Xcode’s built-in Git support has improved significantly over the past few releases, it still falls short in both usability and functionality for many everyday workflows. Developed by Karthik K, Gitwyn is a native Git client designed around the Xcode workflow. Press ⌘⇧E in any Xcode project to review changes, organize staged files, generate commit messages, and push code—all without leaving Xcode or switching to another Git tool.
One of Gitwyn’s standout features is its ability to generate commit messages locally using on-device Apple Intelligence based on staged changes. No network connection is required, and your source code never leaves your Mac. More importantly, it streamlines the entire flow of reviewing diffs, organizing commits, generating descriptions, and pushing changes, keeping Git operations within the Xcode development context.
VersionedDocC: Bringing Version Management to Swift-DocC
For years, every WWDC I would immediately open Apple’s developer documentation and use the “API Changes” feature to compare different Xcode and SDK releases, tracking newly introduced, modified, and deprecated APIs. Unfortunately, this incredibly useful capability disappeared after the documentation website redesign in 2025. Although the old comparison URLs still exist, they no longer present complete API changes.
Developed by Kyle Ye, VersionedDocC brings this capability back to DocC. It adds a release orchestration layer on top of the existing Swift toolchain, generating stable URLs for different versions, providing a version selector, displaying public API differences between adjacent releases based on symbol graphs, and avoiding repeated builds of historical documentation through immutable caching. The project also includes a SwiftPM plugin, GitHub Action, and GitHub Pages workflow, with support for multi-platform symbol graphs, external modules, and standalone DocC catalogs.
Swift Subprocess 1.0
After a year of iteration, Swift’s official swift-subprocess project released version 1.0 this week. It is a cross-platform subprocess library built on Swift Concurrency, providing a unified way to launch commands, supply input, stream output, and manage process termination and cancellation across macOS, Linux, Windows, Android, and BSD.
It is particularly well suited for Swift CLIs, build systems, server applications, developer tools, and macOS automation. Note that version 1.0 requires Swift 6.2 and is not intended for regular iOS applications. It also deliberately leaves higher-level features such as sandboxing, command authorization, and pseudo-terminals (PTYs) to other layers.
swift-subprocess also fills an important missing piece for building native AI agents in Swift: reliably executing external tools, continuously receiving their output, and properly managing subprocess lifecycles. The model decides what to do, while Subprocess provides a more controllable, observable, and Swifty way to execute those decisions.
Thanks for reading Fatbobman’s Swift Weekly! This post is public so feel free to share it.
Apple Intelligence 已通过中国政府审核，即将在中国提供服务
在 WWDC 26 上，苹果公布了新的 Apple Intelligence 架构，不仅在云端引入 Gemini，第三代 Apple Foundation Models 也确认由苹果与 Google 合作构建。这一度让我以为，Apple Intelligence 进入中国市场的难度反而进一步增加了。
但事情的发展比预想中更快。7 月，Apple Intelligence 已经出现在中国国家互联网信息办公室（CAC）公布的最新一批生成式 AI 服务备案名单中。而就在两天前，苹果又确认阿里巴巴 Qwen 将参与中国版 Apple Intelligence，并一度出现在苹果的官方文档中。虽然该文档在 8 月 8 日短暂上线后于次日被暂时撤回，但其中详细的操作指引和隐私说明，已经向外界揭示了相当完整的集成方式。这意味着，Apple Intelligence 在中国正式提供服务已经进入倒计时。
除了阿里巴巴的 Qwen，备案信息还披露，百度也将作为合作伙伴深度参与中国版 Apple Intelligence 的 AI 搜索与视觉搜索。这意味着，苹果不仅可以根据地区调整底层模型和服务提供商，而且这样的合作关系也并非只能由一家厂商承担。
随着各个国家和地区对隐私、数据以及 AI 服务监管要求的不断加强，区域化差异几乎已经不可避免。苹果正在中国率先实践一种新的方式：在保持 Apple Intelligence 整体产品体验和开发接口相对统一的同时，根据不同地区的法规与生态选择不同的底层模型。未来，这种模式很可能还会出现在更多市场，并成为 Apple Intelligence 全球化过程中相当重要的一部分。
对于使用 Apple Intelligence API 的开发者来说，这同样是一个好消息。如果苹果能够继续维持 API 层的一致性，开发者理论上无需为了不同市场维护完全不同的实现方案，也意味着那些以 Apple Intelligence 能力为重要特色的应用，将有机会触达更广泛的用户。
Gemini 和 Qwen 或许并不是我日常 Agent 使用场景中最耀眼的两个名字，但它们在世界知识、语言理解等方面都有相当不错的表现。考虑到 Apple Intelligence 当前更偏向系统级辅助能力，而这些合作伙伴又拥有成熟的模型基础设施和云服务经验，苹果这样的选择其实相当合理。
Apple Intelligence 已经成为苹果在 AI 时代最重要的入口之一。它目前的能力距离我的个人期待仍有一些距离，但至少从 2026 年开始，它正在真正覆盖全球更多用户。相比于今天它究竟能完成多少任务，我更期待 WWDC 2027 上，苹果会如何进一步让 AI 深入融入自己的硬件、系统与服务生态。
如果您发现这份周报或我的博客对您有所帮助，可以考虑通过 Buy Me a Coffee 支持我的创作。
原创
在 SwiftUI Text 中控制孤行：挖掘未公开的 avoidsOrphans
SwiftUI 的 Text 有一项默认行为：当段落末行只剩一个孤词或孤字时，它会主动把上一行的完整词组一并推到末行，避免“落单”。这份体贴多数时候是对的，但代价是上一行可能留下一大片空白——在窄容器或中英混排的场景中，反而破坏了段落的均衡。麻烦在于，UIKit 允许开发者通过 NSParagraphStyle 关掉这项策略，SwiftUI 却没有提供任何公开接口。本文将挖掘 SwiftUI 中一个早已存在、却始终未曾公开的 API —— avoidsOrphans，让开发者重新拿回这项控制权。
近期推荐
再谈协议与全局 Actor 的设计权衡 (An Even Closer Look at Protocols and Global Actors)
随着 Swift 并发模型不断演进，一些过去被视为理所当然的工程取舍，也开始因语言能力的增强而发生变化。在本文中，Matt Massicotte 并没有停留在讨论“@MainActor 应该放在 protocol 上还是方法上”，而是借此进一步探讨了 Swift 并发时代的 API 设计思路。相比给出一种所谓的“最佳实践”，Matt 更强调一种设计思路：隔离是一种约束，应当源于真实需求，而不是沿袭既有习惯。
Tuist 每周动态第一期 (Weekly Tuist News #1)
随着 AI Agent 越来越深入开发流程，Tuist 这类工程管理工具的重要性也在不断提升，对于独立开发者而言同样如此。
作为长期活跃于 Tuist 社区的开发者，Lee YoungJun 推出了一个专门追踪 Tuist 生态的新系列 Weekly Tuist News。与单纯汇总 release notes 不同，这个系列还会结合社区动态以及作者在实际项目中的使用体验。在创刊号中，他介绍了 SwifterPM 成为默认依赖恢复方式、Target 缓存失效条件的新控制能力，以及 Tuist 面向 Coding Agent 的 auth.md 支持，并结合实际使用分享了自己的验证过程与观察。
使用 #Preview(arguments:) 轻松预览 SwiftUI 视图的多状态
大多数 SwiftUI 组件都不只有一种状态：空数据、加载中、错误、权限受限等。开发过程中，我们通常会维护一组重复的 #Preview，或将所有状态放进一个 VStack 中统一展示。Xcode 27 新增的 #Preview(arguments:) 正是为这一场景而设计，它能够根据一组状态数据自动生成多个独立的 Preview，并支持在 Canvas 中单独查看每一种状态。
Artem Novichkov 在文中也分享了他组织 Preview 的思路：arguments 应该描述业务状态（State），而不是环境（Environment）。即使未来 API 继续演进，这种将业务状态与运行环境分离的设计理念依然具有长期参考价值。
用 iOS 27 新 API 实现 SwiftUI 同心圆角按钮 (Concentric Buttons with OS 27’s SwiftUI APIs)
WWDC 26 中，苹果给 SwiftUI 新增了 GeometryProxy.concentricCornerRadii，让开发者能够获取容器的同心圆角信息。不过，ButtonBorderShape 仍然是一个封闭类型，无法直接实现真正的自定义按钮边框。Alexander Weiß 利用这一新 API，结合 onGeometryChange、自定义 ViewModifier 与方法重载，实现了一套使用方式几乎与系统 API 无异的同心按钮方案。
对于
GeometryProxy.concentricCornerRadii这类新增 API，我们往往只需几秒钟就能知道它的作用；但真正应用到项目中时，才会发现还有不少实现细节和框架限制需要考虑。与其说这是一篇 API 介绍，不如说 Alexander 分享了它在实际使用中的注意事项，以及值得尝试的应用场景。
Swift、AI 与软件开发的未来展望 (Swift, AI, and the future of software development)
AI 已经成为许多开发者工作流中不可或缺的一部分，但它在提升效率的同时，也悄然改变着开发者与代码之间的关系。Tibor Bödecs 在这篇文章中真诚分享了自己这一年来的困惑：一方面享受 AI 带来的高效率，另一方面又怀念亲手写代码、深入理解问题的过程；既相信 AI 会持续改变软件开发，也希望它始终只是帮助开发者思考与创造的工具，而不是取代思考本身。
在过去一段时间中，AI 给我带来的困惑与 Tibor Bödecs 很相似。尽管它让我失去了一些乐趣，对代码、对产品产生了某种疏离感，但我也一直不懈地尝试对 AI 建立更强的掌控力，让它更好地为我所用，成为我的外脑，而不是让我沦为它的辅助。
工具
Gitwyn：围绕 Xcode 工作流设计的 Git 客户端
尽管近几个版本中 Xcode 的 Git 功能已经有了明显改善，但无论是使用体验还是功能完整性，仍难以满足部分开发者的在不同场景下的需求。由 Karthik K 开发的 Gitwyn，是一款围绕 Xcode 工作流设计的原生 Git 客户端。在任意 Xcode 项目中按下 ⌘⇧E，即可查看改动、整理暂存内容、生成提交信息并推送代码，全程无需切换到终端或其他 Git 工具。
Gitwyn 的一项特色功能是利用设备端 Apple Intelligence，根据已暂存的代码自动生成提交信息。无需联网，也不会上传源码。相比单纯减少一次输入，它更希望将查看差异、组织提交、生成说明和推送代码整合为一段连续的开发流程，让 Git 操作尽可能留在 Xcode 的工作上下文中。
VersionedDocC：为 Swift-DocC 补上版本管理
过去几年，每逢 WWDC 召开，我通常都会第一时间打开苹果开发者文档，通过“API 变化（API Changes）”功能比较不同 Xcode 与 SDK 版本，查看 API 的新增、修改与弃用。遗憾的是，这项颇为实用的功能在 2025 年的文档网站改版后消失了。原有的差异查询地址虽然仍可访问，却已无法呈现完整的 API 变化。
Kyle Ye 开发的 VersionedDocC，恰好为 DocC 补上了这一能力。它在原有 Swift 工具链之上增加了一层发布编排：为不同版本生成稳定地址，提供版本选择器，依据符号图展示相邻版本的公开 API 变化，并通过不可变缓存避免反复构建历史文档。项目同时提供 SwiftPM 插件、GitHub Action 和 GitHub Pages 工作流，也支持多平台符号图、外部模块及独立 DocC 目录。
Swift Subprocess 1.0
历经一年迭代，Swift 官方项目 swift-subprocess 本周发布 1.0 正式版。它是一套基于 Swift Concurrency 的跨平台子进程库，可以在 macOS、Linux、Windows、Android 与 BSD 系统上，以统一方式启动命令、提供输入、流式读取输出，并管理进程的退出与取消。
它尤其适合 Swift CLI、构建工具、服务器程序、开发者工具与 macOS 自动化。需要注意的是，1.0 要求 Swift 6.2，不面向普通 iOS App，也未提供沙箱、命令授权或伪终端（PTY）等更高层能力。
swift-subprocess 也补上了 Swift 构建本地 Agent 所需的一块重要拼图：可靠执行工具、持续接收运行结果，并妥善管理子进程生命周期。模型决定“做什么”，Subprocess 则让“如何执行”变得更可控、更可观测，也更符合 Swift 的语言范式。