pgrust Project Rewrites PostgreSQL in Rust, Passes Full Regression Test Suite

·6 views

pgrust is an experimental open-source project that aims to completely rewrite the PostgreSQL database engine in Rust, replacing its decades-old C codebase. The project recently reached a significant milestone by passing 100% of the official PostgreSQL 18.3 regression test suite, covering over 46,000 test queries. Unlike wrappers or extensions, pgrust is a ground-up reimplementation that remains disk-compatible with existing PostgreSQL 18.3 database directories. It replaces PostgreSQL's memory-heavy process-per-connection model with a thread-per-connection approach, with early benchmarks showing up to 300x faster performance on analytical workloads. A small engineering team accelerated development using AI coding agents, producing over 250,000 lines of Rust code, though the project is not yet production-ready.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer builds automated WinPE deployment script to replace manual DISM commands

A developer at a company where third-party imaging tools are restricted turned the limitation into an opportunity by building a custom Windows deployment script from scratch. The solution, called deploy.cmd, consolidates disk partitioning, image application, and bootloader configuration into a single self-contained Batch script running within WinPE. To eliminate manual filename entry errors, the script dynamically scans a directory for .wim files and presents them as a numbered menu using delayed variable expansion. Diskpart instructions are generated on-the-fly into a temporary file, executed silently, and logged for troubleshooting, while bcdboot automatically configures UEFI boot files after imaging. The complete scripts have been published on GitHub for others facing similar enterprise environment restrictions.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Builds Free AI-Powered RFP Compliance Checker Using Groq and Cloudflare

A developer built a browser-based tool that checks procurement bid responses against specification requirements, flagging compliant, partial, and missing clauses. The tool uses Groq's free-tier API running the Llama 3.3-70B model for inference, with a single Cloudflare Worker serving as the backend. No user accounts, databases, or stored data are involved, keeping ongoing infrastructure costs at zero. The most challenging aspect was engineering a prompt that reliably returns structured JSON across varied procurement document formats, solved by combining a strict JSON schema, explicit response formatting, and a near-zero temperature setting. The developer plans to release additional single-purpose procurement tools in the same series, with a plain-English bid-grader announced as the next project.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Rust Borrowing Explained: Immutable vs Mutable Access and Why It Matters

Rust's borrowing system controls data access through two forms: immutable borrows for read-only use and mutable borrows for exclusive write access. The core rule is straightforward — multiple readers are allowed, or one writer, but never both simultaneously. This design prevents common bugs such as accidental mutation, stale data reads, and race conditions that are frequent in languages like Python, Java, or C. The borrow checker enforces these access patterns at compile time, making code behavior predictable without requiring runtime checks. Rather than a restriction, Rust's borrowing model is intended as a safety-focused design tool that encourages developers to think clearly about data ownership and access lifecycles.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Chinese AI Models Match US Rivals on Performance at a Fraction of the Cost

An AI infrastructure architect who routes production traffic across three continents has published 30 days of real-world data comparing US and Chinese large language model APIs. Pricing data shows Chinese models like DeepSeek V4 Flash cost as little as $0.25 per million output tokens, compared to $15.00 for Anthropic's Claude 3.5 Sonnet — a 60-fold difference that can translate to hundreds of thousands of dollars in savings at scale. Latency monitoring from multiple global regions found Chinese models performing comparably to US counterparts, with p99 response times mostly under 1.2 seconds when routed through a proper multi-region gateway. Benchmark scores across reasoning, code generation, and Chinese-language tasks showed Chinese models trailing US leaders by only one to three percentage points in most categories. The author concludes that for high-volume workloads, the cost gap between US and Chinese AI providers has grown large enough to materially affect product profitability.

0 comments Read more at DEV Community