每日新聞摘要

The author built Folio, a free Android file viewer after finding existing apps bloated with ads, sign-ins, and broad permissions. Its defining rule is that it requests no INTERNET permission at all, and it also avoids storage permission by using Android’s Storage Access Framework for user-granted access to files and folders. Folio supports many document, media, and archive formats, while editing is limited to text-like files and Office documents remain view-only. The post also explains the offline-first architecture: local crash reports and email-based feedback, update checks handled in the browser, and a GitHub Releases distribution model because it is not shipped through the Play Store.

作者在受够了带广告、强制登录和过度权限的文件查看器后，自己做了一个免费的 Android 应用 Folio。它最关键的规则是完全不申请 INTERNET 权限，也不要求存储权限，而是通过 Android 的 Storage Access Framework 访问用户明确打开或授权的文件与文件夹。Folio 支持 PDF、Office、文本、图片、音频/视频和 ZIP 等多种格式，并刻意让 Word、Excel、PowerPoint 只提供查看功能。文章还介绍了离线反馈、崩溃报告、更新检查，以及因为不通过 Play 商店而采用 GitHub Releases 发布的方式。

One week after launching FinLedger on Google Play, the author found that real users quickly validated—and challenged—his assumptions. The strongest response came from people relieved to use a finance app without mandatory sign-ins, proving that privacy and immediate access are meaningful advantages. At the same time, users made it clear that offline-first cannot mean being trapped on one device, so FinLedger offers encrypted local backups, private Google Drive sync, and open CSV/PDF exports. Another widely shared pain point was credit card double-counting, where card swipes and bill payments are mistakenly treated as expenses. The post closes by framing public shipping as the start of continuous refinement, not the end, with upcoming work focused on safer editing, richer statements, and on-device SMS intelligence.

FinLedger 在 Google Play 上线一周后，作者从真实用户那里验证并修正了许多假设。最直接的反馈是：大家很欢迎一款无需强制登录、即可立即使用的财务 App，这说明隐私和即时启动本身就是重要优势。同时，用户也提醒他，offline-first 不能意味着数据只能困在单一设备中，因此 FinLedger 提供加密的本地备份、私有的 Google Drive 同步，以及可导出的 CSV 和 PDF。另一个被广泛讨论的问题是信用卡重复计算：刷卡与后续还款不应都被算作支出。文章最后指出，公开发布只是持续打磨的开始，接下来会聚焦更安全的编辑、更多周期报表和设备端 SMS 智能分析。

The author compares building the same list feature with offline cache and pull to refresh in Flutter, Swift, and Kotlin. Flutter’s challenge was not the UI itself, but learning to keep state out of the widget tree so rebuilds stay cheap. SwiftUI felt elegant until navigation state introduced awkward enums and paths, and iOS review exposed untested screens. In Kotlin, coroutines made asynchronous code pleasant, but lifecycle management became the trap: a scope attached to the wrong owner can leak in ways that are difficult to reproduce. Across all three, the real feature was not the list, but cache invalidation and empty states.

作者在 Flutter、Swift 和 Kotlin 中分别实现了同一个带离线缓存和下拉刷新的列表功能。Flutter 的难点不在 UI，而在于理解 widget 重建只有在你把它做得足够便宜时才真的便宜，因此要把状态从树中移出去。SwiftUI 起初写起来很漂亮，但导航状态会让代码变成 enums 和 paths 的组合，iOS 审核还可能发现你没测试到的页面。Kotlin 里 coroutine 让异步部分更顺手，不过生命周期处理容易踩坑，绑定到错误 owner 的 scope 可能造成难复现的泄漏。真正的功能不是列表，而是缓存失效和空状态。

Purple TV is an unofficial Twitch client built for Android TV. In the author’s 2017 Sony TV setup, it can play 1080p60 video at 60 fps, contrasting with the official Twitch app’s weaker performance on the same device. The article positions Purple TV as a custom solution tailored to improve playback on older TV hardware.

Purple TV 是一款面向 Android TV 的非官方 Twitch 客户端。作者提到，在他的 2017 年 Sony 电视上，这个应用可以以 60 fps 播放 1080p60 视频；相比之下，官方 Twitch 应用在同一设备上的表现较弱。文章主要说明，通过自定义方案，较旧的电视硬件也能获得更流畅的播放体验。

The article introduces an Android file viewer built without the internet permission, reflecting the author’s frustration with existing apps that request broad storage access, insert ads between pages, or require sign-in just to open a file. The excerpt positions the project as a response to those common app behaviors, emphasizing privacy and simplicity without claiming additional features or performance details.

这篇文章介绍了一款没有互联网权限的 Android 文件查看器，源于作者对现有应用的不满：它们常常要求访问整个存储空间、在页面之间插入广告，或者为了打开文件就要求登录。该内容将这个项目描述为对这些常见问题的回应，重点放在隐私与简洁上，并未声称额外功能或性能优势。

Both snapshotFlow and derivedStateOf observe Compose state and react to changes, but they are not interchangeable alternatives. The article’s core point is that the difference between them is not based on what state they watch. Instead, each serves a distinct purpose in Compose, so choosing between them depends on the behavior you need rather than assuming one can replace the other.

snapshotFlow 和 derivedStateOf 都会观察 Compose 状态并对变化作出响应，但它们并不是可以互相替代的方案。文章的核心观点是：区分它们的重点不在于“观察哪些状态”，而在于各自承担不同用途。在 Compose 中应根据实际需求选择，而不是默认其中一个可以直接替代另一个。

Kotlin Weekly #535, published on November 1, 2026, highlights recent Kotlin ecosystem news, including the Shipaton 2026 winners, Kotlin 2.5.0-Beta2, the State of Kotlin in 2026 report, and KotlinConf 2027 in Kraków. It also points to Android developer tooling updates from Android Dev Summit, such as new Android CLI commands for agents, Compose recomposition tracing, and R8 in AGP 9.3, plus Android Studio Rabbit 2 becoming stable with Bring Your Own Agent. The issue rounds out with Multiplatform articles, sponsored learning resources, and a Libraries section featuring early-stage compiler, formatter, calendar, scheduler, and AI-related projects.

Kotlin Weekly #535 发布于 2026 年 11 月 1 日，汇总了近期 Kotlin 生态新闻，包括 Shipaton 2026 获奖者、Kotlin 2.5.0-Beta2、《State of Kotlin in 2026》报告，以及将在克拉科夫举行的 KotlinConf 2027。内容还介绍了 Android Dev Summit 带来的工具和性能更新，例如面向 agents 的 Android CLI 新命令、Compose 重组追踪，以及 AGP 9.3 中的 R8，并提到 Android Studio Rabbit 2 已稳定及 Bring Your Own Agent 功能。除此之外，还有 Multiplatform 相关文章、赞助资源和一组仍处于早期阶段的库项目。

The author explains why Muffle was built as an offline-first Android app instead of relying on Firebase or another cloud backend. The goal was to automate sound profiles based on prayer times, meetings, and geofenced locations while keeping sensitive data on the device. Room stores routines and geofence rules locally, and GeofencingClient triggers AudioManager actions without sending coordinates to a server. This approach improves privacy and works without internet, but it also makes reliability harder: reboot recovery, background service persistence, WakeLock handling, and OEM-specific battery restrictions all required extra engineering. The article concludes that local-only design can reduce security and compliance burden for privacy-sensitive utility apps, though it shifts more responsibility onto the developer.

作者解释了 Muffle 为什么选择 offline-first 的 Android 架构，而不是依赖 Firebase 或其他云端后端。应用需要根据礼拜时间、会议和地理围栏位置自动调整铃声模式，因此他决定把敏感数据保留在设备上。Room 以本地方式保存例程和 geofence 规则，GeofencingClient 触发后直接驱动 AudioManager，无需把坐标发送到服务器。这样的设计提升了隐私，也能在没有网络的情况下运行，但代价是系统重启后的任务恢复、后台服务持久性、WakeLock 处理，以及不同 OEM 的省电限制都需要额外应对。文章最后指出，对隐私敏感的工具类 App 来说，纯本地设计可以降低安全与合规负担，但也把更多可靠性责任转移给开发者。

Android AppFunctions was announced as “officially in beta,” but the Jetpack library behind it is still 1.0.0-alpha10, and the docs continue to call it an experimental preview. The feature lets an app act like a local MCP server: annotated Kotlin functions are indexed by the OS and can be invoked by an agent inside the app’s own process, without a network hop. But it is gated by Android 16, compileSdk 36, and the EXECUTE_APP_FUNCTIONS permission, while real-world reach remains limited. The article argues that KDoc is now part of the interface, because those comments are what the agent reads, and that most teams should treat this as a learning exercise rather than a product decision—for now.

Android AppFunctions 被宣布为“正式进入 beta”，但其对应的 Jetpack 库仍是 1.0.0-alpha10，文档页面也依旧标注为 experimental preview。这个功能让应用充当本地 MCP server：OS 会索引带注解的 Kotlin 函数，agent 可以在应用进程内直接调用，而不经过网络。与此同时，它受 Android 16、compileSdk 36 和 EXECUTE_APP_FUNCTIONS 权限限制，现实中的覆盖范围仍然很窄。文章的核心观点是：KDoc 不再只是文档，而是 agent 读取的工具说明；对大多数 Android 团队来说，现在更适合把它当作学习和试验，而不是立即做成产品决策。

Part 11 explains how an Android AI agent can recover from a failed step without restarting the whole task. Instead of regenerating the entire plan, the system sends the LLM the completed steps, the failed step, the failure reason, and a compact snapshot of the current screen, then asks for revised steps only for the remaining work. The article shows how to serialize screen state into a concise prompt, merge the revised steps back into the task while preserving completed work, and decide when not to replan, such as repeated failures on the same step or confirmed actions that still fail.

第 11 部分介绍 Android AI agent 在步骤失败后如何不必重来，而是只为剩余任务重新规划。系统会把已完成步骤、失败步骤、失败原因以及当前屏幕快照交给 LLM，请它只生成后续的新步骤。文章还展示了如何将屏幕状态压缩成简洁提示、如何把修订后的步骤合并回任务，以及哪些情况不应该重新规划，例如同一步骤连续失败，或用户已确认的动作仍然失败。

FinLedger found that its Jetpack Compose transaction list was dropping frames because expensive work was running on the main thread, multiple Room Flow queries were emitting independently, and unstable models were forcing unnecessary recompositions. The fix was a broader pipeline refactor: offload list transformations to Dispatchers.Default, debounce search input, merge eight flows into three shared hot streams, switch to collectAsStateWithLifecycle, and define Compose stability rules for core models. After these changes, scrolling became fluid at 60/120fps, startup improved, and memory spikes during list updates were reduced.

FinLedger 发现，Jetpack Compose 的交易列表之所以掉帧，是因为大量列表运算都在主线程上执行，多个 Room Flow 又分别触发 emission，而不稳定的模型类则导致不必要的重组。团队通过重构数据管线，把分组、排序和格式化迁移到 Dispatchers.Default，加入搜索 debounce，将 8 个 Flow 整合为 3 个共享热流，改用 collectAsStateWithLifecycle，并在 compose_stability.conf 中声明核心模型的稳定性。结果是滚动更流畅，启动更快，列表更新时的内存尖峰也下降了。

The author introduces Plavio, a streaming app for Android phones and Android TV built as a small ecosystem of components. The core app provides the interface and player, while PlavioHD handles movie, series, and live TV streams, PMeta supplies posters, cast, ratings, and metadata, and Plavio Search speeds up title discovery. The app also supports additional addons such as players, subtitles, downloads, and debrid services. A key focus is usability on slow networks: PlavioHD checks and caches streams on the server side so playback starts quickly in the app. Plavio is free to use, ad-free, and shares the same account and library between phone and TV apps.

作者介绍了自己开发的 Plavio，这是一款面向 Android 手机和 Android TV 的流媒体应用，由多个组件组成。核心应用负责界面和播放器，PlavioHD 处理电影、剧集与直播流，PMeta 提供海报、演员、评分和完整信息，Plavio Search 则帮助快速找到标题。作者还说明可以接入其他插件，例如播放器、字幕、下载和 debrid 支持。Plavio 重点考虑慢速网络场景，PlavioHD 会在服务器端检查并缓存流，使应用内几乎能立刻打开播放；同时它免费、无广告，手机和电视应用共用同一账号与媒体库。

This article briefs readers on the first Android Dev Summit since 2022, focusing on what Google announced, how mature each part of the platform is, and how to apply it in practice. It is aimed at Android developers who want a concise overview of the summit’s key takeaways, with code examples included to help explain usage. The piece emphasizes both the announcements themselves and the readiness level of each item, rather than treating everything as fully stable or complete.

本文梳理了自 2022 年以来首届 Android Dev Summit 的重点，聚焦 Google 发布了哪些内容、各项功能目前成熟到什么程度，以及在实践中该如何使用，并配有代码示例。文章面向 Android 开发者，提供一份简明的大会摘要，特别强调各项能力的成熟度与使用方式，而不是把所有内容都视为已经完全稳定。

The article explains a field operations app being designed for places with unreliable internet. The author says they are building the delivery engine before the screens, prioritizing the underlying workflow over the interface. A key idea is saving work on the device first and then receiving acknowledgment from the server, highlighting a sync approach suited to intermittent connectivity. The piece frames this as a fundamental architectural choice rather than a UI-first development process.

这篇文章介绍作者正在为网络不稳定的地点打造一款外勤作业应用。作者表示自己会先构建传送引擎，而不是先做界面，重点放在底层工作流而非前端呈现。核心思路是先把工作保存到设备上，再等待服务器确认，体现出适合间歇性连接场景的同步设计。文章强调这是一项架构层面的选择，而不是以 UI 为中心的开发方式。

Google has shipped two alpha libraries under the AppState name, and the accompanying guide demonstrates Compose State being used in repositories, services, and other layers. The article raises the question of whether ViewModel is becoming obsolete as AppState takes shape. Since the libraries are still in alpha, the piece is best read as an early look at a proposal rather than a finished replacement. It highlights the direction Google is exploring without claiming that AppState has already fully replaced existing architecture patterns.

Google 以 AppState 之名发布了两个 alpha 库，并在配套指南中展示了 Compose State 可以在 repository、service 等多个层级运行。文章因此提出一个问题：随着 AppState 出现，ViewModel 是否正在变得过时。不过这些库仍处于 alpha 阶段，所以更适合把它看作早期提案或探索，而不是已经完成的正式替代方案。文章主要传达的是 Google 正在尝试的方向，而非宣称 AppState 已全面取代现有架构模式。

The piece presents a short analogy: a model is like a tenant, and the room should be built before it is advertised. It emphasizes preparing the underlying structure first rather than promoting something before it exists. The provided text is very brief and appears to be an excerpt that directs readers to continue on Medium, so no further technical details are given.

这段文字用一个简短的比喻来表达观点：model 就像 tenant，应该先把 room 建好，再去宣传。核心意思是先准备好底层结构，再对外展示或推广，而不是在基础尚未完成时就急着发布。原文内容非常简短，并以“Continue reading on Medium”结尾，因此没有更多技术细节。

This article compares `TextFieldState` and `TextFieldValue` in Jetpack Compose by rebuilding the same login form twice. The author argues that using `BasicTextField(state)` can eliminate asynchronous cursor-jumping bugs, make ViewModel-based validation simpler, and remove a significant amount of boilerplate code. The piece focuses on practical form handling trade-offs for Compose developers, especially around state management and text input behavior.

本文通过两次重做同一个登录表单，对比了 Jetpack Compose 中的 `TextFieldState` 与 `TextFieldValue`。作者认为，使用 `BasicTextField(state)` 可以消除因异步带来的光标跳动问题，让基于 ViewModel 的校验逻辑更简单，也能减少不少样板代码。文章主要讨论 Compose 表单处理中的实际取舍，尤其是状态管理和文本输入行为。