I’ve recently set up a Bluesky Personal Data Server (PDS) to store the data for my Bluesky account. I wanted to host it on my server alongside the many other web apps, databases, and many other services. I additionally wanted to use my top-level domain as my handle.

I started out following the install guide on the official PDS repo and it initially started out pretty well. However, I pretty quickly ran into some issues where the default config didn’t work for me.

The Problem

I was working on a screen-space reflection shader. This involves reading pixels out of the main scene’s framebuffer/texture to be reflected.

I was seeing strange grainy-looking artifacts showing up in my reflections that I couldn’t trace to anything in my shader code. Here’s what they looked like:

Triplanar Mapping is a method I make use of all the time in my 3D projects. Over time, I’ve experimented with tweaks and alterations to it with the goal of making it look better, run more performantly, and be useful in more situations.

The main change I present here is switching from using a linear mix of texture weights from each axis to a non-linear mix.

To explain what I mean by this, I’ll just show some code.

I regularly use Graphviz to generate… graph visualizations. Since my website and blog are all dark-themed, I usually want to set it up so that the generated SVG is dark-themed as well.

Here’s the basic config I use to make graphviz produce outputs with dark backgrounds and light content:

digraph G {
  bgcolor="#181818";

  node [
    fontcolor = "#e6e6e6",
    style = filled,
    color = "#e6e6e6",
    fillcolor = "#333333"
  ]

  edge [
    color = "#e6e6e6",
    fontcolor = "#e6e6e6"
  ]

  # The rest of your code goes here...
  A -> B;
  B -> C;
  C -> A;
}

That results in an output that looks like this:

Recently when working with the regl WebGL wrapper library, I encountered a weird error that I didn’t understand the cause of:

(regl) invalid dynamic attribute "position" in command

I checked my code, and I was indeed passing a position attribute and the data array I was using to construct the buffer looked correct. The error also happened intermittently, happening after my application hot-reloaded.

The Cause

It turns out that I had accidentally created two regl instances and was trying to use buffers allocated on one with commands created on a different one.

I was working on a new SvelteKit project with Svelte 5 recently, and I kept getting an error that prevented anything from loading in the browser:

[vite] Pre-transform error: Failed to load url __SERVER__/internal.js (resolved id: /Users/casey/osu-embeddings/frontend/.svelte-kit/generated/server/internal.js) in /Users/casey/osu-embeddings/frontend/node_modules/@sveltejs/kit/src/runtime/server/index.js. Does the file exist?
[vite] Error when evaluating SSR module /node_modules/@sveltejs/kit/src/runtime/server/index.js: failed to import "__SERVER__/internal.js"
|- Error: Cannot find module '__SERVER__/internal.js' imported from '/Users/casey/osu-embeddings/frontend/node_modules/@sveltejs/kit/src/runtime/server/index.js'
    at nodeImport (file:///Users/casey/osu-embeddings/frontend/node_modules/vite/dist/node/chunks/dep-DkOS1hkm.js:55067:25)
    at ssrImport (file:///Users/casey/osu-embeddings/frontend/node_modules/vite/dist/node/chunks/dep-DkOS1hkm.js:54976:30)
    at eval (/Users/casey/osu-embeddings/frontend/node_modules/@sveltejs/kit/src/runtime/server/index.js:5:37)
    at async instantiateModule (file:///Users/casey/osu-embeddings/frontend/node_modules/vite/dist/node/chunks/dep-DkOS1hkm.js:55036:9)

The Cause

It turns out that I had a broken symlink in my static/ directory. I had cloned the project down to a new computer, and the file only existed on the old one.

The Problem

I recently created a new SvelteKit project using the Svelte 5 preview and kept getting an error at the beginning of every Svelte component:

The error was e.children.findLastIndex is not a function (svelte). It was showing up even on the basic SvelteKit skeleton starter project as soon as I added a <style></style> block.

I run self-hosted Plausible analytics to keep track of how many people are visiting my various websites and web apps. I’m largely very happy with it - it gives me all of the info I need with only a miniscule JS payload, no cookies or other invasive tracking, and full control over the data.

However, recently I’ve been running into an issue with fake/bogus traffic getting submitted for my personal homepage. Usually, I’m more than happy to see a bump in traffic to one of my sites, but something was off about it this time.

For a new project at my dayjob, I’ve had the opportunity to try out sea-orm for the database layer. In the past, I’ve tried out other Rust SQL solutions including diesel and sqlx, so I have some context to compare this one to.

At a high level, sea-orm provides a fully-featured solution for managing your database setup in Rust. It provides a framework and CLI for setting up and maintaining migrations, code-gen’ing entities and relations, and writing + running queries. Like most other Rust DB options, it is fully typed and integration into Rust’s type system.

I recently received a bug report on a library I built - three-good-godrays - which implements screen-space raymarched godrays for Three.JS as a pass for the pmndrs postprocessing library. One of the problems pointed out was that colors seemed washed out/desaturated when my pass was used, even when the pass wasn’t rendering any godrays.

Here’s how things look by default without the effect (and are supposed to look with it on):