Remco Haszing

Remco Haszing

@remcohaszing.nl

Open source software developer, cat servant, coffee to code converter, metalhead 🤘

957 followers257 following1.4K posts

Remember when LLMs could supposedly understand everything based on just the content? Why do we have all those LLM related config files now?

Remco Haszing reposted

Can you give me a genuine reason for someone to use a site/app in a color mode (light/dark) different from their current OS mode? So far, the only argument I head was "well this site has a garbage light/dark mode". Okay, say the site has a perfect light/dark modes.

Netflix used an LLM to translate a kids show. The voice overs were so bad, it was considered harmful to child development. After some heat, they removed the Dutch version. I saw this in Dutch media. I don’t think that really justifies it. They deserve international backlash over this.

rtl.nlKinderserie vol taalfouten offline gehaald door NetflixNetflix heeft de populaire kinderserie Spike offline gehaald na ophef over de nasynchronisatie van het programma. De streamingdienst had daarvoor gebruikgemaakt van kunstmatige intelligentie, in plaat...

Can we all agree that slopforking is just unethical and disrespectful to the original authors? Doesn’t matter if it’s from source or reverse engineered. Competition is good, but so is maintenance and acknowledgement. Reverse engineering is an impressive feat, but not if it’s without effort.

Does anyone have any recommendations for hosting code sandboxes? Our Stackblitz embeds are broken. CodeSandbox doesn’t work either.

Remco Haszing reposted

We have $100K to give away to open source maintainers! Let us know your suggestions and nominations.

Remco Haszing reposted

The Transloadit API is now machine-readable: a public OpenAPI 3.1 spec, webhooks included, for typed clients and Postman imports, a JSON Schema to validate Assembly Notifications, and every docs page as Markdown plus llms.txt, so your coding agent reads the real docs.

transloadit.comA machine-readable API: OpenAPI 3.1, Markdown docs, and llms.txtGenerate typed Transloadit clients from our OpenAPI 3.1 spec, validate webhooks against real schemas, and give AI agents every docs page as Markdown.

Cool! My first security advisory from a third party for one of my personal projects! Oh wait… it’s LLM spam from a new GitHub account

LLMs are tools, not people. This means they are not authors either. They are not the author of your git commits, you are. It’s good to disclose if you used an LLM to make edits, but the commit author field is not the right place. That’s just a weird place to put an ad for the tool.

This also means you deserve both blame and attribution for the change. And it’s also your job to give proper attribution if the code was copied from somewhere else.

I made some significant changes to mdx-content-mapper. It now uses micromark tokens instead of mdast. This makes the mappings much more stable and gives better handling if certain syntax errors. But my cat won’t allow me to move off the couch, so I won’t make a release today.

Remco Haszing reposted

This is such a silly mistake and yet I see you making it all the time in your tests. ❌ Do NOT run cleanup after assertions ✅ Schedule the cleanup via hooks or use Disposable Objects A seemingly tiny change that'll have tremendous impact on your tests, especially memory-wise.

Why do people use the "prepublishOnly" and "prepare" npm script? I feel like they serve no real purpose, but people use them because they’re used and written about a lot.

JavaScrip quiz: Which lines contain function declarations? 1: const a = function() {} 2: const b = function c() {} 3: const d = () => {} 4: export default function() {} 5: export default function e() {} 6: export default () = {}

Can characters be escaped in code blocks in JSDoc comments? I want something like this: /** * ```js * /* This is a comment inside a code block inside a comment */ * ``` */ It only has to work for TypeScript, not every JSDoc tool out there.

Remco Haszing reposted

I'm declaring TypeScript 7 for Ember: ready At work, on the 3 million line monolithic package, I'm seeing ~ 4m40s -> 12s a ✨ 23x ✨ speed boost

`assert()` style test assertions are so much better than `expect()`. I really don’t mind the stylistic choice. But `assert()` style assertions help TypeScript narrow down types, whereas `expect()` style assertions can’t do that.

How do people feel about showing 2 LSP definitions for 1 character? I really like how definitions are working out for headers, strong, etc. But in case of markdown lists, the <ul> is defined by the same token that creates the <li>

This is what it looks like if you define invalid overrides for the "ol" / "ul" components. For an invalid "li" override, every list item would be marked as an error.

Remco Haszing reposted

webpack 5.110.0 🎉 ⚡ New performance hints suite 🎨 Native CSS is the way forward — css-loader, style-loader & mini-css-extract-plugin deprecated 🗜️ Built-in CSS & HTML minification 📦 nodeModules externals preset 🌳 Better tree shaking & scope hoisting webpack.js.org/blog/2026-08...

webpack.js.orgWebpack 5.110 | webpackwebpack is a module bundler. Its main purpose is to bundle JavaScript files for usage in a browser, yet it is also capable of transforming, bundling, or packaging just about any resource or asset.

I’m now seeing type definitions, hover info, type errors, etc, for h1-h6 headers in MDX that are overridden in mdx-components.tsx This uses the TypeScript content mapper Atom span maps. I believe Volar was never even be able to do this. But also, I never tried that. I might have overlooked it.

View full thread

I guess you can use this to forbid specific element types by providing something like null as a component override. That will be a type error now. Though I think this would be more of a job for a linter.

I’m now seeing type definitions, hover info, type errors, etc, for h1-h6 headers in MDX that are overridden in mdx-components.tsx This uses the TypeScript content mapper Atom span maps. I believe Volar was never even be able to do this. But also, I never tried that. I might have overlooked it.

I could do this for emphasis, strong text, inline code, etc. What’s going to be interesting, is markdown constructs that yield multiple elements, such as lists and GFM tables.

I just released the final version of the Volar based @mdx-js/typescript-plugin and the last version of the language server and VSCode extension that will include this. Newer versions will use the new TypeScript 7.1 content mapper.

Remco Haszing reposted

Ember now has preliminary TS7 support ~ via the content-mappers feature that merged in TS last week Testers wanted! (requires TS Nightly)

I added props inference for @nextjs.org in mdx-content-mapper. Also type information is inferred from mdx-components.tsx (or any providerImportSource of your choice). I could also infer types for known exports, but maybe that’s taking it too far?