每日新聞摘要

Ferran Pons's article "How to Run LLMs Offline on Android Using Kotlin" outlines the concept of executing large language models locally on Android devices using Kotlin. It emphasizes on-device AI, a privacy-first approach, and removing dependence on remote servers. The piece frames local inference as a way to keep user data on-device, reduce latency, and maintain functionality without network access. Although brief, the article positions on-device deployment as a practical, privacy-centered alternative to cloud-hosted models and points readers toward further reading on Medium. The topic is relevant for developers evaluating serverless AI options for mobile applications and considering resource trade-offs such as model size, device compute limits, and potential hardware acceleration for efficient inference.

Ferran Pons 的文章《How to Run LLMs Offline on Android Using Kotlin》介绍了如何在 Android 设备上使用 Kotlin 本地运行大型语言模型（LLM）。文章强调 on-device AI、privacy-first 与 no servers required，说明设备端推理能将用户数据保留在本机、降低延迟并在无网络时仍可工作。文章以概念性说明为主，并引导读者到 Medium 上继续阅读，适合评估移动端无服务器 AI 方案的开发者，同时提示需关注模型大小、设备算力与加速选项等实际部署考量。

In Jetpack Compose, the order in which modifiers are applied can have a large and often overlooked impact on rendering performance while leaving the visual output unchanged. Sergey Drymchenko demonstrates that a simple reorder of modifiers can transform the same UI from janky to butter‑smooth without changing any pixels. Because modifiers are executed sequentially and can affect measurement, layout, drawing, clipping, and layer creation, their arrangement changes the amount and type of work Compose performs during composition and recomposition. Reordering can avoid redundant layout passes, reduce extra graphic layers, and minimize expensive operations, leading to better frame timing and responsiveness. The article recommends using profiling tools to spot performance hotspots and experimenting with modifier order as a low-effort optimization technique that can yield substantial improvements without altering the UI’s appearance.

在 Jetpack Compose 中，Modifier 的应用顺序会显著影响渲染性能，但不会改变最终的像素显示。Sergey Drymchenko 指出，仅通过简单地调整 Modifier 排列，就能让相同的界面从卡顿变得非常流畅。由于 Modifier 是按顺序作用于测量、布局与绘制，顺序会改变 Compose 在组合和重组时需要执行的工作量。通过合理重排可以避免不必要的布局计算、减少额外图层或昂贵操作，从而改善帧率和响应性。文章建议使用性能分析工具找出热点，并以 Modifier 顺序调整作为一种低成本、高回报的优化方法。

KotlinConf brings developers from around the world together to learn from leading experts, explore the latest industry insights, and connect with the Kotlin community. KotlinConf’25 welcomed attendees from over 50 countries. With KotlinConf’26 only a few months away, the organizers have announced the first round of speakers set to appear on stage—talented individuals described as shaping the future of the language and its broader ecosystem. The upcoming conference promises inspiring talks, detailed workshops, and a range of opportunities to learn, exchange ideas, and network with fellow Kotlin enthusiasts. The announcement invites interested attendees to secure tickets for an unforgettable conference experience and notes that additional speakers and the full schedule will be released soon.

每屆 KotlinConf 都吸引世界各地的開發者前來向專家學習、掌握最新業界見解並與社群互動；KotlinConf’25 有來自超過 50 個國家的與會者。距離 KotlinConf’26 只剩幾個月，主辦方已公布首批登台講者，他們被視為塑造語言與生態系未來的重要人物。會議將包括啟發性的演講、深入的工作坊和多樣的交流機會；同時呼籲有興趣者儘早購票，並表示更多講者與完整議程將陸續公布。

In this tutorial Aleksandar Pelanovic explains idempotence for Kafka producers, how it prevents duplicate records by assigning a unique Producer ID and sequence numbers, and how that behavior affects throughput. He outlines required settings (acks=all, retries>0, max.in.flight.requests.per.connection<=5) and notes that idempotence is default since Kafka 3.0 unless conflicting options are set. Using JMH and Docker Compose, benchmarks compare idempotent and non‑idempotent producers on a single broker and a three‑broker cluster (replication factor three), with careful batching and timeouts to stabilize results. Throughput measurements (~24,000 ops/s) show only minor differences between modes and run variability can exceed those differences. The article concludes idempotence provides strong deduplication with little performance penalty—keep it enabled for correctness, and disable it only when duplicates are acceptable and absolute peak throughput is paramount.

本文介绍了 Kafka 生产者的幂等性概念、实现方式及其对性能的影响。作者解释了如何在协议层通过唯一 Producer ID 与序列号识别并丢弃重试产生的重复记录，并列出了确保幂等性的必要配置（acks=all、允许重试、max.in.flight<=5），同时指出 Kafka 3.0 起在无冲突配置时默认启用幂等性。使用 JMH 和 Docker Compose，在单节点与三节点（副本因子 3）的环境下进行基准测试，结果在约 24000 条/s 的吞吐水平上显示启用与未启用幂等性的差异很小且在测量波动范围内。结论为：若需避免重复并保证投递可靠，建议启用幂等；仅在可接受重复且追求极致吞吐时可考虑关闭。

Eclipse's automatic formatter enforces consistent code style but can harm readability for specialized blocks such as multi-line SQL, manual tables, or ASCII diagrams. This article explains how to disable formatting globally via Preferences → Java > Editor > Save Actions, and how to protect specific code sections by enabling and using formatter off/on tags (for example, @formatter:off and @formatter:on) under Java > Code Style > Formatter. It also covers finer techniques: enabling Never join already wrapped lines to preserve manual line breaks, using trailing comments to prevent line merging, and renaming tags to match team conventions. Best practices include using off/on tags only when necessary, ensuring teammates enable the tags to avoid accidental reformatting, and testing settings with Ctrl+Shift+F before committing to version control.

Eclipse 的自动格式化有助于保持风格一致，但在多行 SQL、手工对齐或 ASCII 图等场景会破坏可读性。本文介绍如何通过 Preferences → Java > Editor > Save Actions 在全局禁用自动格式化，以及如何在 Java > Code Style > Formatter 中启用并使用 @formatter:off / @formatter:on 标记来保护特定代码区块。还描述了更细致的控制方法：启用 Never join already wrapped lines 保留手动换行、用尾注（//）防止行被合并，以及自定义标记名称以符合团队约定。建议仅在必要时使用 off/on 标记，确保团队成员都启用该功能，并在提交前用 Ctrl+Shift+F 测试设置以避免版本控制中的格式差异。

This article surveys practical ways to run TestNG tests selectively to avoid rerunning an entire suite when only a few tests fail. TestNG auto-generates testng-failed.xml (in target/surefire-reports) after a run that captures failed and skipped methods, allowing direct reruns. You can also craft a testng.xml with <include> entries to execute chosen methods, or invoke a single method from Maven Surefire using -Dtest=Class#method. For dynamic control, throw SkipException from @BeforeMethod to skip specific methods at runtime, or implement IMethodInterceptor to filter or reorder discovered methods before execution. IDEs like IntelliJ offer quick, no-configuration single-method runs for debugging. Each approach has trade-offs: testng-failed.xml and Maven suit CI reruns, manual testng.xml is ad hoc, while @BeforeMethod and IMethodInterceptor centralize selection logic for flexible debugging and reduced setup overhead.

本文概述了在 TestNG 中选择性执行测试的实用方法，避免在少数测试失败时耗费资源重跑整个套件。TestNG 在执行完成且存在失败或跳过的测试时，会生成 testng-failed.xml，仅包含失败/跳过的方法，可直接用于重跑。也可以手动编写包含 <include> 的 testng.xml，或用 Maven Surefire 的 -Dtest=Class#method 指定单个方法执行。对于运行时控制，可在 @BeforeMethod 中抛出 SkipException 跳过特定方法，或实现 IMethodInterceptor 在发现阶段过滤方法。IDE（如 IntelliJ）支持零配置的单方法执行与调试，适合快速迭代和定位故障。

This tutorial explains how to configure proxy authentication in Java across four common HTTP clients: the built-in Java 11+ HttpClient, Apache HttpClient 5, Spring's synchronous RestTemplate, and Spring's reactive WebClient. It introduces a ProxyConfig helper that encapsulates proxy host, port, username and password and provides methods for a JVM Authenticator, ProxySelector and Proxy object to reduce boilerplate. For each client it shows how to supply proxy details and credentials: HttpClient uses proxySelector() and authenticator(), Apache HttpClient 5 uses a BasicCredentialsProvider with UsernamePasswordCredentials (char[] password), RestTemplate configures a SimpleClientHttpRequestFactory and JVM-wide Authenticator.setDefault(), and WebClient configures Reactor Netty's HttpClient with ProxyProvider username/password. The article concludes by summarizing when each client is appropriate based on project needs.

本文介绍如何在 Java 中为需要认证的代理服务器配置代理认证，涵盖 Java 11+ HttpClient、Apache HttpClient 5、Spring RestTemplate 与 Spring WebClient 四种常见方案。先定义 ProxyConfig 封装 host、port、username 和 password，并提供 JVM Authenticator、ProxySelector 与 Proxy 实例以减少样板代码。随后分别说明：Java HttpClient 通过 proxySelector() 与 authenticator() 自动处理 407 握手；Apache HttpClient 5 使用 BasicCredentialsProvider 与接受 char[] 密码的 UsernamePasswordCredentials；RestTemplate 以 SimpleClientHttpRequestFactory 配合 Authenticator.setDefault()（JVM 全局）；WebClient 则在 Reactor Netty HttpClient 上配置 ProxyProvider 的主机、端口与用户名/密码。结尾根据项目需求建议不同的客户端选择。

This article demonstrates how to access Samba (SMB) shares from Java using the codelibs JCIFS implementation (SMB3). It shows the Maven dependency to include JCIFS and walks through a test Samba server set up on an Ubuntu VirtualBox guest with a host-only adapter and a fixed IP (192.168.56.101). The guide creates two shares—an anonymous publicshare and a password-protected sambashare—configures permissions and users, and presents code examples. Examples cover the default anonymous context, authenticated context via NtlmPasswordAuthenticator and withCredentials, and common SmbFile operations: exists(), listFiles(), isDirectory(), length(), lastModified(), createNewFile(), delete(), mkdir()/mkdirs(), and copyTo(). It also explains using SmbFileInputStream/SmbFileOutputStream to transfer local files to Samba shares. The article concludes by summarizing setup, authentication, file manipulations, and stream-based I/O with JCIFS.

本文介绍如何使用 codelibs JCIFS（支持 SMB3）在 Java 中访问 Samba 共享。示例在 VirtualBox 的 Ubuntu Server 虚拟机上配置 host-only 网络，并将 DHCP 地址范围限定为单一 IP（192.168.56.101），创建用户 jane，并建立两个共享：匿名的 publicshare 和需密码的 sambashare。文章包含 Maven 依赖、smb.conf 配置段与命令，并演示如何使用 SingletonContext 作为默认上下文、通过 NtlmPasswordAuthenticator 与 withCredentials 获取认证上下文以列出受保护共享内容。随后展示 SmbFile 的常见操作（exists、listFiles、isDirectory、length、lastModified、createNewFile、delete、mkdir(s) 与 copyTo），并说明使用 SmbFileInputStream/SmbFileOutputStream 将本地文件流式传输到 Samba。

The article argues that developers often default to using WebSocket for real-time, event-driven user interfaces when a simpler, already-available tool might suffice. It suggests considering Firebase Cloud Messaging (FCM) as an alternative for pushing events to clients, highlighting that the most advanced technology is not always the best choice if it increases complexity and maintenance. By evaluating requirements and leveraging existing infrastructure like FCM where appropriate, teams can reduce implementation overhead while still delivering responsive interfaces. The core message is to choose the right tool for the job rather than assuming WebSocket is the default solution.

文章提醒开发者不要默认使用 WebSocket 构建事件驱动的界面，因为已有的、更简单的工具在很多场景下同样能满足需求且减少复杂度。作者建议在适当情况下使用 Firebase Cloud Messaging（FCM）向客户端推送事件，以降低实现与维护成本。核心观点是根据实际需求评估技术选型，优先使用已存在且易于集成和维护的方案，而非盲目追求最先进的技术。

In this follow-up, Daniel revisits the Koin dependency injection (DI) framework, recalling that in 2020 it stood out as an alternative to Dagger and offered a simpler developer experience. The article focuses on how Koin has evolved to help scale projects by adopting annotations and the Kotlin Symbol Processing (KSP) toolchain, and by integrating with Navigation 3. Rather than deep benchmarking or exhaustive migration guides, the piece highlights practical paths for modernizing DI setup with Koin’s newer features and interoperability points, aiming to guide developers who liked Koin’s simplicity but need stronger tooling and navigation support for larger Android projects.

作者 Daniel 对 Koin 依赖注入框架进行了回顾，回忆到 2020 年 Koin 作为 Dagger 的替代方案，以更简洁的开发体验受到关注。文章重点在于 Koin 如何借助注解与 Kotlin Symbol Processing（KSP）来支持项目扩展，以及如何与 Navigation 3 集成以满足导航需求。内容并非详尽的性能对比或迁移指南，而是提供实用路线，帮助偏好 Koin 简洁性的开发者在需要更强工具链与导航支持时，了解可行的现代化做法。