Memory64 WebAssembly breaks 4GB heap limit but reveals JS-Wasm ABI mismatch

·1 views

A developer testing WebAssembly's Memory64 feature found it could allocate over 473 million fp32 model parameters in 3.7 seconds, far exceeding the ~250 million parameter ceiling imposed by the traditional 32-bit WebAssembly heap. The 32-bit module was unable to complete the same allocation, confirming Memory64's expanded theoretical ceiling. However, the benchmark was conducted outside the browser environment, leaving the JavaScript-to-WebAssembly bridge untested. A type mismatch emerged at that boundary, where pointer values were passed as Number in some cases and BigInt in others, causing larger browser presets to fall back to the 32-bit path. The developer concluded that while Memory64 raises the allocation limit, an ABI integration issue still blocks a fully functional in-browser workflow.

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)

12-Year-Old Developer Battles Hosting Failures to Keep AI Tutoring App Live

A 12-year-old developer faced back-to-back hosting crises on the launch day of his online festival for KODA, a self-built AI coding mentor app. Netlify blocked new deployments after free build credits ran out, prompting a switch to manual ZIP uploads to get version 16 live. A move to Neocities as a permanent host then revealed a second problem: the platform's Content Security Policy prevented the app's JavaScript from reaching the external AI provider. However, pre-built safeguards — including a 15-second API timeout, a Supabase error-logging table, and a graceful offline fallback — meant users saw a clean message rather than a broken interface. The developer plans to migrate to Cloudflare Pages with a proper GitHub integration to avoid similar hosting dependencies in the future.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Optimistic Locking in Laravel Prevents Data Corruption in Concurrent Systems

High-concurrency applications face a 'Lost Update' problem when multiple processes modify the same database record simultaneously, causing data to be incorrectly overwritten. Two common approaches exist to handle this: pessimistic locking, which blocks records during transactions but risks bottlenecks, and optimistic locking, which allows concurrent reads while verifying data integrity only at the moment of update. A technical guide published on DEV Community outlines how Laravel developers can implement optimistic locking using a PostgreSQL version column that increments on every update. The approach involves creating a reusable Eloquent trait that automatically appends version checks to update queries, eliminating the need for manual intervention across controllers. This method is presented as better suited for modern, high-traffic web applications where availability and performance are priorities.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Preserve Invoice PDFs as Immutable Evidence in Dispute Workflows

A technical guide published on DEV Community outlines best practices for handling invoice PDFs during billing disputes in Node.js applications. The core advice warns against re-rendering old invoices using current templates, since factors like metadata, font subsets, and compression can silently alter file bytes even when output looks identical. Instead, developers are urged to store original PDF files alongside a SHA-256 hash, template version, fonts, and rendering tool details to create a verifiable evidence package. The guide provides TypeScript code that validates a stored PDF's hash before use and deliberately refuses to auto-correct any mismatch, treating byte-level changes as new facts requiring review. When originals are unavailable, reconstruction must be clearly labeled as such and kept strictly separate from verified original copies.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Solves in Hours What AI Tools Failed to Fix Over Two Weeks

A developer spent nearly two weeks trying to reorganize an ESP-IDF project into a components folder, relying on free and paid AI tools including Gemini, DeepSeek, and GitHub Copilot, none of which produced working results. Each AI-generated solution triggered widespread errors in the code editor, and repeated attempts across multiple models yielded the same outcome. After stepping away and returning with a fresh approach, the developer stopped blindly applying AI suggestions and began manually editing key configuration files. This hands-on review revealed that all the AI tools had consistently used an incorrect package name in the REQUIRES field of the CMakeLists.txt file. A single Google search uncovered the correct name, and fixing that one line resolved the entire issue immediately.

0 comments Read more at DEV Community