每日新聞摘要

Java 9+ strong encapsulation and JPMS can prevent Gson from using reflection on private fields, causing InaccessibleObjectException during deserialization in modular projects. The article shows that a POJO may fail on the module path, while a Java record works because Gson uses its canonical constructor. For POJOs, the article presents four approaches: make fields public, open the package to com.google.gson, write a custom TypeAdapter, or use the JVM --add-opens argument as a last resort. It also explains that Gson 2.14.0 restores support for nested java.time values through built-in adapters, while ISO-formatted date strings still require a custom adapter.

Java 9 之后的强封装与 JPMS 会限制 Gson 通过反射访问私有字段，因此在模块化项目中反序列化时可能抛出 InaccessibleObjectException。文章展示了 POJO 在 module path 下会失败，而 record 因 Gson 会使用其 canonical constructor，所以可以直接成功反序列化。若仍需使用 POJO，可采用四种办法：将字段设为 public、使用 opens 将包开放给 com.google.gson、实现自定义 TypeAdapter，或在启动时加入 --add-opens 参数。文中还说明 Gson 2.14.0 已恢复对嵌套 java.time 数据的支持，但 ISO 格式日期字符串仍需要自定义 adapter。

The author split a single-module Android app into a core/feature module structure while keeping the same functionality and screens, then compared incremental build times by touching one file and rebuilding. The monolith build took 3.744 seconds, while the modularized version took 2.614 seconds, about 30% faster. The repository includes both the main and monolith-baseline versions so others can reproduce the comparison. The stack used is Kotlin, Compose, Hilt, and Gradle version catalogs.

作者将一个单模块 Android 应用拆分为 core/feature 模块结构，在功能和界面完全相同的前提下，通过修改一个文件并重新构建来比较增量构建时间。结果显示，单体版本耗时 3.744 秒，模块化版本耗时 2.614 秒，约快 30%。仓库同时包含 main 和 monolith-baseline 两个版本，方便读者自行复现。使用的技术栈包括 Kotlin、Compose、Hilt 和 Gradle version catalogs。

This article explains how a shared e-bike operations app was built around a Kotlin Multiplatform core instead of duplicating the product for Android and iOS. The team first analyzed the app’s roughly 40 screens and found that most were stable business forms, while only a few areas truly needed native SDKs such as maps, Bluetooth, camera scanning, location, and uploads. Based on that split, they chose KMP for business logic, Compose Multiplatform for shared UI, H5 for dashboards, and thin host shells for platform-specific capabilities. Two migration rounds then moved map screens and the navigation shell out of Android host code, raising shared code from 81% to 95.8%. The article also covers i18n, platform contracts, and the repo’s current shipping scope.

本文介绍团队如何以 Kotlin Multiplatform 为核心，构建一套 Android 与 iOS 共享的电助力车运维 App，而不是为两端分别重复开发。作者先分析约 40 个页面，发现大多数都是稳定的表单、列表和状态流转，真正需要原生能力的只有地图、Bluetooth、相机扫描、定位和上传等少数部分，因此采用 KMP 承担业务核心、Compose Multiplatform 实现共享 UI、H5 承载数据看板，并用轻量宿主注入平台能力。随后通过两轮迁移，把地图页面和导航壳层逐步移出 Android host，使共享代码比例从 81% 提升到 95.8%，同时也说明了国际化、平台接口和当前已交付的功能范围。

This final installment in the seven-part series presents a case study of rendering Jetpack Compose directly in a terminal using Mosaic. The article’s core point is that Compose can function as a UI-agnostic runtime, not just as a toolkit for traditional Android screens. By targeting the terminal, the series aims to demonstrate Compose’s flexibility across presentation layers while staying within the constraints described in the original post.

这是七篇系列的最后一篇，以 Mosaic 为例，说明如何将 Jetpack Compose 直接渲染到终端中。文章要表达的核心观点是：Compose 不只是用于传统 Android 界面的工具，也可以作为与 UI 无关的 runtime 使用。通过把输出目标放到 terminal，系列文章用这个案例展示了 Compose 在不同呈现层上的灵活性。

The article argues that many Android interview prep lists are shallow and unhelpful because they focus on textbook definitions instead of real-world engineering concerns. In actual senior interviews, candidates are more likely to be asked about practical issues such as why an app drains the battery, rather than simply defining core Android components. The piece sets up a more realistic view of interview preparation for Android roles.

文章指出，很多 Android 面试准备清单过于肤浅，往往只要求背诵 Activity 等基础概念，对真正面试帮助不大。作者认为，资深 Android 面试更常考察的是实际工程问题，比如应用为什么会耗电，而不是单纯测试对基础知识的记忆。本文借此强调，准备方向应从教科书式定义转向真实场景下的问题分析与解决能力。

The provided excerpt is extremely limited and includes only the heading, author, link, and a brief introduction note that says “Continue reading on Medium.” It does not contain any actual discussion of Jetpack Navigation 3, Compose changes, feature updates, previews, or implementation details. As a result, there is no substantive technical content to summarize beyond the fact that the full article is hosted on Medium.

这段原文摘录非常有限，只包含标题、作者、链接，以及一句“Continue reading on Medium”。摘录中没有真正介绍 Jetpack Navigation 3 在 Compose 方面的更新、功能变化、预览状态或实现细节，因此无法从现有内容整理出具体技术重点；能够确认的信息只有完整文章需到 Medium 查看。