Nibbles, Bits & Bytes.

Woke up to a massive “god candle” appearing in my Umami instance for Plant🌱Smart. I had to do a triple-take as I immediately wrote it off as Google finally re-indexing the site after almost 2 years of nothing. Then, I took a much closer look and noticed 99% of the traffic was coming from email clients. Which, to be frank, I found quite odd. Why email clients? As it turns out, The Morning Brew — a popular news… newsletter in the States — mentioned my little project in that morning’s issue. Almost 10k visits to the site when the most that could be expected was, maybe, 20 per day if I was lucky. After all, I haven’t really done much in the way of advertising it. The big spike has obviously fizzled out quite a bit, but as a result, the site has been linked to and mentioned in several other places. This can only be good for the site’s rankings in Google’s index, so no complaints coming from me!

I’m obviously a little bummed out the traffic has steadied out as much as it has, but this is a small purpose-built utility site. I’m not going to get 1,000’s of visitors per day here as a normal occurrence. Just stoked that people find it interesting!

From the maker themselves, “OKpalette is a perceptual color extraction tool that generates beautiful color palettes from images using advanced color space analysis.” Such a cool little “small web” project. Love seeing useful stuff like this out in the wild.

I’ve been rebuilding the public landing page for the homelab this weekend and somewhere in the middle of it I laid a scanline over the whole thing. It’s a single hairline that repeats itself every three pixels. It’s faint enough that the page stops looking quite so flat without you ever working out why. Then, obviously, I wanted it here too.

The lazy version is a white line at about 1.5% opacity, which is what the other page uses, because the other page is forty words of neon over black; legibility isn’t really the goal. However, this site is mostly prose, so a white line lightens the words as well as the background. I didn’t spend all that time on the reading column just to lay a haze over it.

So the grain mixes off --text instead of white:

--grain: color-mix(in srgb, var(--text) 1.5%, transparent);

This can be applied like so:

.grain {
  position: fixed;
  inset: 0;
  pointer-events: none;
  z-index: 10;
  background: repeating-linear-gradient(
    to bottom,
    var(--grain) 0 1px,
    transparent 1px 3px
  );
}

And finally referened like so just above the closing </body> tag:

<div class="grain" aria-hidden="true"></div>

This works out great, because a colour mixed with itself gives that colour straight back. The grain physically cannot touch the body copy. I’ve learned quite a few new things today!

As a bonus the lightbox is a <dialog> and the browser promotes a modal dialog into the top layer. This sits above every z-index on the page; including the one holding the grain. So a zoomed photograph comes up perfectly clean and I didn’t have to write a rule to make that happen.

Whether you can see any of this depends entirely on your monitor, your colour settings and even your resolution. If you can’t see it, it’s working.

Page loads had been creeping up again, so I went looking. Turns out I’d been quietly serving enormous screenshots. One post was pulling 5.1 MB of PNGs, and a single one of those was a 1.96 MB app screenshot sitting in a column 736 pixels wide. Oof.

The fix was mostly plumbing. My img shortcode was handing the raw src straight to the <img> tag, so Hugo never got a look in. Everything now goes through one partial that resizes to the width it actually displays at, re-encodes to WebP and emits a srcset, plus width and height so the layout stops jumping around while things load. Markdown images get the same treatment through a render hook. The awkward bit: images under static/ aren’t resources, so Hugo can’t touch them. A second mount in hugo.toml fixes that without breaking any existing URLs.

While I was in there I dealt with the JavaScript. Every page was carrying all seven inline scripts, even the 90-odd pages with no code blocks, no zoomable images and no locked series titles. Those three now load only where they actually do something. The two that genuinely run everywhere moved into one small fingerprinted bundle, so your browser fetches it once for the whole visit instead of re-reading it on every page. I also finally turned on HTML minification, which I’d apparently never bothered with.

Where it landed:

That last one is about 46 times smaller. Nothing changed visually, which was the whole point. The sparks still spark and the locked titles still scramble.

One last silly thing. I added a new short code to style “Plant Smart” as “Plant🌱Smart”. Fun!

With all the frenetic redesigning of Plant🌱Smart these past few weeks, I couldn’t believe I completely forgot about the image gallery on the details page. The old design looked a bit dated like everything else, soooo the galleries also get a refresh and lightbox. Yay.

This is still very early days, but Sifa is a professional network built on the AT Protocol. It’s meant to be an open alternative to LinkedIn which, let’s be honest, has always been a bit of a hellsite. Even more so lately with all the AI boosterism on the timeline. Probably may never grow a network as large here as I have on LI, but good riddance. I’ll still be keeping the entry in the side rail navigation for the time being, but its days are numbered.

Welp, I finally did it! I’ve moved the git repository for this site from Github to the Forgejo instance in my homelab. Further, I’ve replaced Cloudflare’s build pipeline with a simple 20-line YAML file used by my Woodpecker CI server, which is also hosted in my homelab. I’m still using Wrangler to push the static assets to Cloudflare Pages and I’ll always need an external DNS provider for my domain, but you can’t reasonably self-host everything.

It’s nice to have almost full control of my little website now! And all it took was 3 attempts to get the integration right. 😅

A great note to end on for a Monday evening!

A measured and reasoned critique of the shit-show that is Omarchy. Classic security engineer snark met with some Shakespearean levels of verbosity. This isn’t necessarily firing shots at the increasingly far-right DHH, but plays into his surrounding cult of personality and how it’s used to lure Linux beginners into a false sense of security based purely on vibes… and a shit tonne of Bash scripts. Credit to the author for avoiding Omarchy’s creator’s politics.

Goodbye Tailwind. Goodbye Node. We’re rolling vanilla HTML, CSS and JS. No more webfonts, we on them system faces now. A couple of hours on I have not missed the webfonts once. The site is lighter, the text is sharper on my screen and the only person who ever noticed was me. 750kb cold start for the main page is now under 18kb.

Everything is so much faster now I should’ve done this months ago when I first revamped the site.

It’s an oldie, but I always go back to it. The post I keep sending to people who tell me Go error handling is just noise. The point was never that you type if err != nil a lot. The point is that an error is an ordinary value, so you can do ordinary things with it.

Spent an evening on the tags. A few of them were doing the same job under different names, so the duplicates are merged and the thin ones are gone. The cloud in the rail is shorter and rather more honest for it.

Read this for the third time and it still lands. Complexity very, very bad. Grug is right more often than I would like to admit.

Opened the bill expecting a fright and found a NAT gateway I have not needed since March. It has been quietly charging me rent to forward packets nowhere. Coding is my passion.

I keep promising myself I will write more on the weekends. The drafts folder keeps returning a different answer.

Sat down to fix one broken keybind. Three hours later the whole tmux config is rewritten and the keybind is still broken.

An interesting read a mate of mine shared about how relatively simple it is to exfiltrate your sandboxed Claude memories when pairing it with an agent that can access the web. Good stuff!

Strangely sudden news today. Looks like “Go Report Card” has been completely sunset as of a couple weeks ago. I really dug that little service. At least it’s still available for self-hosting, but now I have to go on cross-repo badge cleanup duty.

Having a play with a more dynamic menu drawer. Just working out the kinks on smaller screens.

Decided to introduce a new projects page to showcase some of the bigger things I’m working on. I’m still working out the kinks and I’m not 100% sold on the current design. Each card is powered by a .toml based datasource in Hugo, contains a kind of linear heatmap of activity related to the associated project’s tag as well as the 5 latest related posts. Good enough to publish, I guess.

I also can’t decide on which activity graph animation I like the most. 😅

I am very smart. I only just realised over the weekend that issues and discussions in the parent repo for this site were disabled. Meaning, article coments using Giscuss was functionally disabled here. WOOPS!

Welp. I’ve convinced myself that Glazier needs Terraform-style variable declaration blocks to validate --var flags. I’m doing it and I will not be stopped. I’m thinking something as simple as the following session block-adjecent top-level spec:

variable "name" {
  description = optional(string)
  type        = required(string | bool | number)
  default     = optional(type)
}

Time to do what I do best: make things more complicated than they have any right to be.

I should update Plant🌱Smart. Thinking it should get a similar treatment to this site. Maybe even exploring moving completely to Hugo instead of a haphazardly-assembled static Svelte site.

Some small incremental updates for the site today:

• The addition of a new bit type for site updates like this one.
• Proper RSS feeds for both long-form articles and bits. Check the footer!
• A quick sweep of the site map to ensure everything maps out properly. There were a few dead links previously.
• You can now filter bits by type! Have a play with it at the top of this page.
• Some minor grammatical updates here and there for some recent content.

Something to note is that the bits filter is purely JS based and assumes all content will be displayed in a single page. This is fine for now, but there may be 100s or even 1,000s of entries on a long enough timeline. At that point, basic text search and filtering that supports pagination may be necessary. But, not today!

Man, some of these router CVEs are laughably bad. Not saying rolling your own router with OpenWRT is a perfect solution… but compared to this it functionally is.

An interesting study on the concept of “cognitive surrender” or, how we simply offload thinking to an LLM and implicitly trust whatever it spits out the other end.

Tempted to drop the custom fonts altogether. They account for well-over half the transfer on a hard refresh.

I am not a smart man. I completely forgot I was using the “Pro” version of FontAwesome and had the entire library committed to this very public repository without any licensing coverage. I mean, the pro version shouldn’t be shared anyway. You may have noticed the associated Github repository has been temporarily set to private and that all commit history as been wiped. Sucks, but I felt this was much more pragmatic than removing multiple specific files from the tree. I am not an expert git surgeon, so that got a hard pass from me.

I’m currently experimenting with some alternative OPEN SOURCE web icons and updating the third-party licensing notice bundled with the repo to cover icons as well as web fonts.

Luckily, only I noticed this issue and immediately worked to address it.

edit: Good-bye, sweet git history!

I’m keeping these stupid test quotes around. I love them. Fight me.