daily.dev · Popular · 实时热榜

How long can the system be unavailable? How much data can we afford to lose? The first answer helps define your Recovery Time Objective: how quickly service must return. The second helps define your Recovery Point Objective: how far back your recoverable data may be. “Immediately” and “nothing” are popular answers until the cost of achieving them arrives. Recovery requirements must be based on business impact, not optimism. #AWSWithoutTheFluff #DisasterRecovery #BusinessContinuity

A revisit of Kent Beck's Extreme Programming Explained, thirty years after XP's creation, arguing its five values, fourteen principles, and 24-26 practices remain relevant amid AI-driven software development. The piece breaks practices into 'more critical than ever' (whole team, team continuity, energized work, real customer involvement), 'controversial' (pair programming, code-and-tests-only, test-first programming), 'common sense', and 'still relevant?' categories, weaving in commentary on AI

Here’s a strange twist in the Linux desktop story: Ubuntu’s fastest-growing user base isn’t running Linux at all. It’s running Windows. According to Canonical’s VP of Engineering, Jon Seager, Ubuntu installs via Windows Subsystem for Linux (WSL) are growing faster than native Ubuntu desktop installs. He expects WSL usage to overtake native desktop usage within months. Soon, most people running Ubuntu could be doing it from inside Windows. Why this is happening A lot of developers don’t get to pi

A long-time Obsidian user shares how they finally made the app stick after two failed attempts, mostly derailed by plugin overload. The breakthrough was mentally decoupling the vault (a folder of markdown files) from the app itself, treating Obsidian as just one of several tools pointed at that vault. Their current workflow uses Notepad for quick edits, AnythingLLM with a local LLM for retrieval chat over the vault via its Obsidian connector, and the Obsidian app itself mainly for its built-in C

Somehow, in 2024, the internet is having a real argument about dark mode toggles. Not dark mode itself. The toggle . Specifically: should it have two states or three? Lea Verou, who works on web standards and clearly has opinions people listen to, says two is enough. Her logic: a persistent, always-visible toggle can flip between light and dark, defaulting to system preference on first visit and storing an override in localStorage after that. Adding a third “System” option, she argues, just adds

A deep dive into caching for system design covering how caches work internally, where copies of data can live (in-process, Redis/Memcached, CDN, browser, database), and the tradeoffs of each layer. It walks through Redis internals (hash table resizing, approximated LRU/LFU eviction via sampling, lazy and active expiry), Memcached's slab allocation and rebalancing history, HTTP caching semantics (no-cache vs no-store, ETags, stale-while-revalidate, stale-if-error), CDN tiered caching, and why dat

Laravel Lock is a third-party package by Md Mahedi Zaman Zaber that scopes distributed locks to Eloquent models and HTTP routes. It offers a fluent builder (Lock::for()->ttl()->acquire()/block()), a HasLocks trait that derives lock keys from a model's morph class and primary key, a lock route middleware that can target a specific route parameter, and cache or database storage drivers with events for acquisition, failure, and release. It requires PHP 8.2+ and supports Laravel 11, 12, and 13, and

A general advice piece arguing that becoming a better programmer isn't about shortcuts, tools, or AI, but about consistent habits: going beyond tutorials to learn fundamentals, experimenting with working code, stepping outside your comfort zone, taking ownership of hard problems, contributing to open source, surrounding yourself with skilled developers, and teaching what you learn. The overarching message is that steady habits compound into real growth over time.

The funniest thing about the current AI hype is that sometimes it feels like we’re all playing the same game: Developer: “We added an AI feature.” Marketing: “Great. Put AI everywhere on the landing page.” Users: “Oh, it has AI? Take my money.” And suddenly a perfectly normal SaaS product becomes an AI-powered revolutionary platform™ . 😂 But here’s the interesting part: is it actually working? We’re reaching a point where “AI-powered” can get attention, improve your chances of making the shortli

Cursor picked quite a day to ship Origin, its own git hosting platform. GitHub had just eaten a multi-hour outage (again), with error rates around 20% across web and API traffic and roughly 50% for archive downloads. Some are calling the timing genius. Others are calling it a coincidence dressed up as strategy. “People are saying Cursor has great timing to launch Origin but given how often GitHub is down these days I don’t find the timing to be too impressive,” mitsuhiko wrote, which is either t

DuckDB v2.0, codenamed "Cyanoptera," is coming this fall with over 10,000 commits since v1.5. Headline additions include a client/server mode via the Quack extension and new CONNECT statement (with remote pushdown to PostgreSQL/MySQL), a first-class VARIANT type for shredded semi-structured data, full trigger support, several SQL dialect additions (NEAREST joins, DML inside CTEs, nested schemas, variable syntax, JSON mutation functions, recursive CTEs with USING KEY), asynchronous I/O throughout

A UI/UX design debate around dark mode toggles argues that a simple two-state light/dark control can express all three intents (light, dark, system) if the default falls back to system preference and an override is stored in localStorage for future visits. The piece endorses Lea Verou's stance that persistent, always-visible toggles should avoid a redundant System option since it adds cognitive load without real benefit, while noting two legitimate exceptions: settings-panel toggles and interfac

AI now produces T-SQL that looks polished, aligned, aliased, and commented, but formatting no longer signals correctness the way it once did. Five realistic query examples show subtle traps: a DELETE that removes far more customers than intended due to a JOIN vs correlated subquery mix-up, a NOT IN query silently returning zero rows because of a NULL, an UPDATE with non-deterministic results from duplicate staging rows, a BETWEEN date range that excludes the last moments of a month, and a TRY/CA

Every startup landing page suddenly looks the same, and developers are done pretending it’s fine. The pattern has a name now, even if nobody agreed on it together: instrument serif headline, gratuitous italics, an em dash or three, a three-word tagline that means nothing, a curl-into-bash install script, and a “trusted by” row of logos that all happen to be from the same YC batch. A satirical site nailed the aesthetic so precisely that it reads less like parody and more like a bug report against

If a request waits for a PDF, email, image resize and three webhooks, the user is paying for work they do not need to watch. Move slow and retryable work to a queue. But a queue needs more than add() : Idempotent processors Retries and backoff Dead-letter handling Job-status visibility Monitoring Safe payloads A queue should make failure manageable, not move failure somewhere nobody checks. What produced the biggest improvement after you moved it to a queue? #NestJS #BullMQ #BackgroundJobs

The Rust standard library now runs cargo-semver-checks in CI to catch accidental breakage of stable APIs, following a history of incidents where unstable methods or soundness fixes accidentally broke stable code (async-std in 2020, BuildHasher's dyn-safety in 2021, ChunksMut's Send/Sync in 2022, and a tokio Windows breakage in Rust 1.94 that required a 1.94.1 point release). The work required teaching cargo-semver-checks to understand Rust's stability attributes (#[stable], #[unstable], const-st

TLDR: OpenAI keeps losing senior leaders while its own data shows a widening gulf between AI power users and everyone else. Meta’s CTO managed to torch employee goodwill by dismissing a question about vacation time as “very dumb.” A wave of surveys landed this week quantifying what a lot of developers already suspected: AI is generating production incidents, tech debt, and burnout faster than it’s generating trust. Oracle and Halo Studios both cut staff even as their underlying numbers looked fi

A workflow for running AI coding agents overnight without waking up to a broken codebase. The core idea: separate thinking from execution. Write structured plans in Obsidian (goal, scope, success criteria, do-not-touch, open questions), stress-test them with a 'grill-me' skill to surface vague or risky assumptions, rewrite them into boring, unambiguous execution plans, hand execution to a different agent than the one that planned, then run a single 'plan of plans' prompt that points at all prepa

An overview of Tiger Style, the coding philosophy behind TigerBeetle, the financial transactions database, exploring its priority order of safety, performance, then developer experience. Covers explicit limits on queues and buffers, fixed integer widths over architecture-dependent types like usize, avoiding dynamic allocation after startup, heavy use of assertions, batching, cache-conscious data layout, single-threaded transaction processing to avoid contention, minimizing data copying in hot pa

I dont like my tokens go waste, so I have been using my ollama tokens with GLM 5.2 and it is fabulous. I thought these models would never catchup claude but to be honest these are really good, I can say 70 percent. There is still 30 percent probably claude is smarter but I wrote a firefox extension and a small game with godot all has been fun I started with kimi k3 but it was too expensive then I was going through the models, I tried kimi coder 2.7 that was also good but when I start using glm 5

NativePHP v4 introduces SuperNative, a rendering mode that compiles Blade components directly into real SwiftUI views on iOS and Jetpack Compose views on Android, with no web view or HTML involved. Screens are now PHP classes extending NativeComponent paired with Blade templates using native elements styled by Tailwind classes. PHP and native code share memory directly, eliminating bridge round-trips, and accessibility features like VoiceOver and TalkBack work natively. Web views remain supporte

Silo is a community-maintained fork of MinIO (previously known as pgsty/minio), created after MinIO changed its license and removed features like the admin console. Silo restores the admin console, documentation, and other stripped functionality, positioning itself as an open alternative for object storage. The first release under the Silo name shipped as RELEASE.2026-08-06T00-00-00Z.

Upgrading a Symfony application to PHP 8.4 is framed as a strategic, incremental process rather than a full rewrite. PHP 8.4 remains well-supported (active support through 2026, security fixes through 2028) and is now the minimum requirement for Symfony 8.1. Key PHP 8.4 features like property hooks, asymmetric visibility, and native lazy objects (which Symfony 7.3+ leverages automatically for lazy-loaded services) can simplify code without forcing a rewrite. Practical migration steps include che

A hands-on test of the newly released Qwen 3.8 27B model running locally on a 16GB VRAM RTX 5060Ti via llama.cpp. The dense hybrid model (linear attention on 48 of 64 layers plus gated attention) supports native multimodal input including video, a 262K context window expandable to 1M tokens, and toggleable thinking mode. Coding tasks like a portfolio site and a 3D racing game were generated but were slow (4.78-6.42 tokens/sec) and required long generation times (over an hour). Benchmark scores a

An ORM skeptic argues that with LLMs now capable of writing boilerplate code, the traditional justification for using ORMs (avoiding tedious SQL mapping code) largely disappears. The author contends that ORM query optimization already requires deep expertise and workarounds, undermining the promise of abstraction, and suggests it may be simpler to have an LLM write parametrized SQL directly rather than fight with an ORM's implicit API, while cautioning about SQL injection risks.

An engineering leader shares why AI-based code review tools like Cursor's BugBot and custom Claude code review skills became too slow, noisy, and expensive (hitting $1000/day in one case) for a team producing thousands of PRs a quarter. The fix that worked: converting team-specific coding rules from markdown guidelines into custom linters, which run in seconds, are deterministic, can run in-editor and on pre-commit hooks, and stop agents from ever pushing bad code. Several example custom lint ru

As AI coding assistants shift software engineering from writing to reviewing code, Go's design philosophy becomes increasingly valuable. The post argues that Go's strict compiler, unified toolchain (gofmt, govulncheck, native fuzz testing), opinionated simplicity, and strong backward compatibility make it uniquely suited for AI-assisted development. Key points include: Go's static type system catches AI hallucinations at compile time rather than runtime; its comprehensive standard library reduce

Despite social media claims that AI has killed coding interviews, CoderPad's State of Tech Hiring report shows technical assessments are up 48% globally since mid-2023, with 43% of companies still running classic algorithmic tests. Companies like OpenAI, Anthropic, and Meta run some of the strictest interviews, with Meta even building an AI-assisted round that candidates report is harder than the classic version. The argument is that interviews were never really about writing code but about judg

Hi Persons, i am making a Rust version of Skyrim with Bevy, inspired by the OpenMorrowind. I want to make it and i will, this will help modders and gamers play this amazing game on Linux, android, windows and i am aiming for Mac too. But recreating the engine isn’t enough for me, i tough to my self. Why not making an modern launcher, make it a mod manager, so gamers can search, install and organize mods. But this is not enough, why not convert the game files? Let’s do that too, convert the files

sqlfmt is a new opinionated SQL formatter inspired by gofmt, implementing the 'river alignment' style from The Art of PostgreSQL. It right-pads all clause keywords (select, from, where, group by, etc.) to the same column at each nesting level, creating a vertical river with expressions flowing to the right. The tool uses a tokenizer rather than an AST for robustness and comment preservation. It ships with a CLI mirroring gofmt's interface (-w, -l, -d flags), editor plugins for Emacs and Vim/Neov

pgrust 0.2 achieves 300x faster analytical query performance than Postgres on ClickBench, even outperforming ClickHouse. The post walks through three key query engine optimizations using Rust code examples: (1) batching — replacing row-at-a-time Volcano model calls with 1024-element batches, cutting time from 1.3s to 480ms; (2) operator fusion — combining sequential scan and aggregation into a single node to eliminate copy overhead, reaching 358ms; and (3) SIMD — using ARM NEON intrinsics to pro