Chameth.com - posts like building-a-new-computer, g15-ram-upgrade, infinite-avatars but not migrating-from-github-to-forgejo, why-you-should-be-using-https

Making a font of my handwriting

2025-08-08T00:00:00Z

Recently I’ve been on a small campaign to try to make my personal website more… personal. Little ways to make it obvious it’s mine and personal, not just another piece of the boring corporate dystopia that is most of the web these days. I don’t quite want to fully regress to the Geocities era and fill the screen with animated under construction GIFs, but I do want to capture some of that vibe.

I’d added some bits and pieces along those lines: floating images in articles now look like they’re stuck to the page with sellotape, related post links have a wavy border that animates when you hover over them, and so on. Next, I wanted to change the heading fonts from a monospace font to something cursive, to resemble handwriting. Less terminal output, more handwritten letter. I couldn’t find one I liked, though. So why not make my own? It can’t be that hard, right?

Failing to do it myself

I set out to try to make the font myself using open source tools. After doing a bit of research, it seemed like the general approach was to create vectors of each character and then import them into a font editor. That seems to mean either Adobe Illustrator and FontLab (if you have too much money) or Inkscape and FontForge (if you like open source). I fall firmly into the latter category, so I grabbed my graphics tablet and opened Inkscape.

I wrote out my first three letters: capital A, B and C. Saved them in Inkscape, and attempted to import them into FontForge. Then I remembered one crucial thing that had slipped my mind: I absolutely loathe using FontForge. It’s a bit like when you open an old version of GIMP and get a bunch of weird looking windows floating all over the place; it feels like you’re fighting against the tool to do even the most basic operations. The difference is I have cause to edit images a lot more than I edit fonts, and GIMP has actually significantly improved their UI over the years.

Here are the rough steps I went through with FontForge:

Launch Font Forge. It shows a weird bit of art in one window, and an open file dialog in another.
I don’t want to open a file, so I close that dialog. The program exits.
Relaunch Font Forge, and realise that within the “Open Font” dialog is a “New” button. Click it.
Get to the standard font-editing UI. Right-click on the “A” looking for a way to import an SVG. Don’t see one.
Click around a bit, exploring the menus. Everything feels a bit off. You can’t open one menu then hover over the next to see its content, like basically every UI toolkit in existence. I think FontForge has eschewed QT and GTK in favour of doing things itself.
Find the “Import” option in the File menu. Hope it’s for a single glyph not the whole font.
A file picker opens. Again it’s all a bit off from normal desktop conventions. Try to resize it, and just get blank grey space at the bottom.
Type the absolute path I want to go to in the text field.
Get a dialog saying “Not a bdf file /home/chris/etc”. Press OK.
Get a dialog saying “Could not find a bitmap font in”. Press OK.
Press Ctrl+L to see if that lets me enter a path. Click everything in the dialog to try to find a way to enter a path. Get annoyed. Give up. Click through folder-by-folder to get to where I want to be.
Get to the folder and don’t see any files. Change the format to “SVG”. Double-click the newly-visible SVG file.
Get a dialog saying “You must select a glyph before you can import an image into it”. Press OK.
The import dialog goes away, having not imported.
Select the glyph in the main tool area, then repeat the File→Import dance.
It’s actually there now! Open the glyph in the editor and see it’s a complete mess of Bézier curves. I can’t click what I want without accidentally moving a handle for an adjacent curve.
Rage-quit.

I’m sure FontForge is less anger inducing once you’re used to it. And you definitely could use it to build a font like this if you had much more patience than me. I’d had enough of death-by-a-thousand-paper-cuts though.

I briefly tried Inkscape’s built-in support for making an SVG font. It annoyed me a lot less, but it’s fiddly: it seemed like each glyph had to be a single path, so you had to convert the glyphs to paths, then merge them correctly. If you merge them incorrectly then the wrong bits of your letters end up filled (like the inside of the ‘B’). Path manipulation is getting towards the limit of my knowledge of vector editing, and it took a bit of trial and error for each letter that had more than a single stroke. I didn’t fancy doing that for every letter.

I’m usually a big advocate of open source, but this was one of those painful times when it feels like it just falls short. Clunky, painful UI and processes where commercial tools just let you get on with your work.

You can exchange money for goods and services

When I’d been looking for open source tutorials, I found many mentions of a closed source, hosted tool: Calligraphr. It has a free version with limitations (no ligatures, no variations, 75 glyphs per font), and a pro version for £8/month. I’d normally balk at the idea of a subscription for this, but they have the perfect answer: you can make a one-time payment, and your account automatically downgrades back to free after a month. It’s not a hidden option, either, it’s the most prominent button on the upgrade page. That made me happy to give them £8 to play around with the service for a month.

Calligraphr works by having you print templates, write out the letters, then scan them in. It does some magical processing to extract the glyphs, provides tools to tidy them up, align them, etc, and then produces a TTF file for you. You can see some of my completed templates here:

Most of the templates I used for the font

Calligraphr has a nice UI to generate the templates, allowing you to select which glyphs to include. I added the “minimal English”, “basic punctuation” and “Ligatures” sets. That gave me four pages to fill out, and I did them all twice. That let me filter out versions that didn’t work well, and have variants for some letters so the font wasn’t too repetitive. Later on, I went back and added some custom ligatures based on blog post titles that didn’t look quite right: “Re”, “To”, “ers”, “ey”, “hy”, “ra”, “re” and “ty”. Ligatures like this help it look more natural: when we write we don’t just stamp out identical letters regardless of their surroundings, instead they will connect to their neighbours, or overlap slightly, or even share a stroke.

I filled these templates in with a Sharpie, as I wanted a fairly informal, scrap-booky look, and it would also give good solid shapes that should be easy to pick out of the template. I scanned them with the “Scan Document” function on my iPhone, and uploaded the PDFs to Calligraphr.

Iterating and tweaking

The Calligraphr UI allows you to preview the font, but I found it a lot more useful to just download a copy and use it on a local copy of my website. That let me test it with real text, and see how it’d look at the different font sizes I use on the site.

The first version was not great. Despite the guidelines on the template, I apparently wasn’t good at sticking to them. Some letters were floating way off the baseline, and some were sunken below. When those opposites met it looked terrible. Fortunately Calligraphr has a pretty easy tool to slide each letter up and down, and scale it up or down if needed, and you can see it next to other letters as you do it. It took a little bit of time to go through all the variants of all the letters, but the next version looked a lot better.

Another tweak I ended up doing was reducing the spacing between letters. The defaults Calligraphr uses are probably good for a blocky font, but I wanted to put the letters close together to give it more of a joined-up look. Again, this is an easy tool to use, you just drag the sides in or out as desired. While these tweaking steps were probably as fiddly as some of the Inkscape steps I refused to do earlier, they’re a lot more rewarding as you see things improving with each one. It’s a lot easier for me to commit time and effort to improving something that’s already working reasonably, than put that time and energy into an unknown.

Later, I noticed that occasionally there would be a huge gap in a title. Not “the kerning is slightly off” but “there’s enough room to park a bus”. It took me a while to figure out what was happening: a couple of glyphs hadn’t been isolated perfectly and had picked up a few pixels from the template lines at the edge of their boxes. That meant the glyph had a width that covered the actual written glyph, a big gap, and then the rogue marks. At first, I fixed this by just adjusting the width, but that left the little pixels floating awkwardly down-sentence. The proper fix was to use the editing tool and simply delete them, and then Calligraphr snapped the width back to what it should be.

These iterations took a while to do, but I just dipped in and out occasionally over the course of a week, so it didn’t actually feel like too much work. I quite enjoy the process of refining things, too.

Result and a surprise

If you’re viewing this post on my website¹, you can see the font in the headers, captions, and a few other places. Here’s how it compares to my actual handwriting:

My handwriting vs my handwriting font

It’s not close enough to forge documents, but I think it definitely gets across my style, and that’s exactly what I wanted. It’s surprisingly legible even at smaller font sizes — I think the weight of the Sharpie helps here — and at £8 and a bit of manual work was a lot more economical than spending days wresting with open source tools.

A few weeks after I put the finishing touches on the font, I got an e-mail from Calligraphr. As my account had lapsed back to the free version, I was no longer eligible for the “server-side backup” feature. So what did they do? They e-mailed me an exported copy! It’s a JSON file with the properties of each glyph and a base64 encoded image. Not only can I re-upload this to Calligraphr if I resubscribe, I can probably hook something up to edit it should I ever need to. I’m blown away by how pro-user Calligraphr’s business practices are. They’re up-front about pricing, don’t try to get you stuck on an auto-renewing subscription, and automatically export your data. It’s like a breath of fresh air compared to the barrage of dark patterns that other websites foist on us. If you want to make this kind of font, I’d definitely recommend them just because of how nice they are.

And I haven’t changed everything since writing this post… ↩︎

Fixing a loud PSU fan without dying

2025-07-30T00:00:00Z

Three months after I built my new computer, it started annoying me. There would occasionally be a noise that sounded like a fan was catching on a cable, but there weren’t any loose cables to be a problem. Over the course of a few weeks, the sound got progressively worse to the extent that I didn’t want to use the computer without headphones on. I measured the sound at 63 dB, which is about the sound of someone talking. That may not sound terrible, but it’s a constant, nasty noise coming from something that sits about 40cm from my head.

After some investigating¹, I identified the PSU fan as the culprit. I have a Cooler Master V750 SFX, which is not super high-end, but wasn’t cheap, either. It shouldn’t be developing issues after three months. Thankfully, it comes with a ten-year warranty, so it should be easy to get sorted, right?

Warranty woes

I looked at Cooler Master’s warranty, and for issues within the first two years you have to deal with the retailer. That would be Amazon in my case. So I looked at Amazon’s information on warranty issues. Their policy is that if it’s more than 30 days since purchase, you have to send it off to a third-party repair center and wait for them to diagnose and try to repair it. Here’s the kicker:

Usually repairs take up to 20 business days (including delivery time), but could take slightly longer

I use the computer for work, have upcoming LAN parties to go to, and generally can’t do without it for an entire month. That’s assuming they reproduce the issue: the PSU fan only turns on when it reaches a certain temperature, so if you just plug it in briefly it won’t exhibit any symptoms.

If I was going to do a warranty return, I’d need to buy a new PSU first to use while this one was away. If I’m spending £100+ on a PSU, then I’d be getting a better one that doesn’t have known issues with the fans. So when the Cooler Master eventually got returned I’d have no use for it. What a waste.

So the warranty is about as much use as a chocolate teapot. Good to know, I guess. If I wanted decent service, I should have probably used a real retailer, not Amazon. Lesson learnt.

Doing some research

If the warranty is useless to me, maybe I should just void it and try to fix it myself. I did some research online. Most of the results went like this:

if you’re not a trained professional you should not be opening up your PSU at all. There’s a lot of current that goes through it which will seriously hurt/main/kill you

it’s time consuming and potentially dangerous

or some other variety of doom and gloom. There were some actually useful results sprinkled in, though. A post on the smallformfactor.net forums details someone swapping out the fan on the 850W version of the same power supply because it was too noisy. It looks like the fan connector in the PSU is a two-pin connector, and they just used an off-the-shelf adapter to convert the standard four-pin case fan connector. It sounded a bit annoying to make it all fit, though.

A similar post on reddit showed basically the same upgrade, but identified the connector as a 2.54mm JST-XH. They crimped their own connector on, which feels a lot better than using a bulky connector. I do have a crimping set for JST connectors, but I absolutely hate doing them because they’re so fiddly and annoying. I’m not sure if I trust one of my crimps inside a power supply, either.

Looking at the pictures in those posts, I identified the fan as an Ong Hua HA9215VH12FD-F00. I searched to see if I could see any specifications for it. I couldn’t. But what I did get back as my first result was another Reddit post, this one titled “Getting rid of a HA9215SH12FD-F00 fan from your PSU”. It says something about the quality of a component when the first result is a post about getting rid of it because it’s annoying².

All of these mods seemed easy enough, and I’m reasonably confident in my ability to both discharge and avoid touching capacitors, so I probably wouldn’t die. I ordered a Noctua NF-A9x14 fan and some pre-made JST-XH connectors and waited for them to arrive.

Upgrade time

The replacement parts arrived one evening, so when I turned the computer off for the night I unplugged it from the mains, held the power button down for a while, and left it overnight for good measure. The next morning I popped open the case, disconnected and removed the PSU, gleefully tore the “warranty void if removed” sticker off, and opened it up.

Removing the fan was simple: unplug the JST connector and undo the screws holding it to the outer panel. I confirmed that my JST connectors were the right size, then started surgery on the Noctua fan. I carefully removed the heatshrink, cut off the normal fan connector, and separated out the four wires. A quick check of their FAQ confirmed the wiring colours: black was ground, yellow +12V, and blue and green used for speed control (which we’re not using in this instance). I clipped the blue and green wires off near the hub so they wouldn’t be in the way, then cut the yellow and black wires down to length, and attached the JST connector with a solder seal:

Old fan, new fan, burnt mouse mat

In the process I discovered my mouse mat doesn’t like being hit with a heat gun. Whoops. The solder seals are great: the solder in the center melts to form an electrical connection, the parts at the end grip the cable to provide mechanical strength, and the whole thing acts like heatshrink and keeps the connection insulated. A lot less fiddly than crimping JST connectors, and a lot smaller and cheaper than an adapter.

I installed the fan and reassembled the PSU. Then undid it all and tried again, as apparently the face the fan mounts to isn’t quite symmetrical and I’d installed the fan upside down. Connected everything back up, plugged the computer in, booted it up, and… nothing. The PSU doesn’t actually power the fan until it hits a certain temperature. I figured putting it under load would probably heat it up, so ran a graphics benchmark a few times. Eventually the fan spun up. Huzzah.

Soon the entire computer will be beige and brown

Peace at last

I measured the sound of the new fan: 43 dB. That’s a decrease of 20 dB. Due to the weird logarithmic properties of the dB scale that no one understands, that means it sounds about 4 times quieter to human ears. Big improvement!

It’s worth pointing out this isn’t a straight one-to-one upgrade. I don’t know the specs of the old fan in terms of airflow, and I’m just assuming that the Noctua will be good enough by virtue of being a decent fan. They also behave slightly differently as the power ramps up: the stock fan starts spinning at a lower voltage than the Noctua, which needs around 7V to get started properly³. I don’t think this will be an issue, as it’s effectively just putting a small step at the lower end of the fan curve. It might shorten the overall lifespan of the PSU, but from my point of view it failed after four months, so any extra time is a bonus.

I recorded some before/after audio, so you can hear the difference. I think there’s some amount of volume normalisation going on, so it’s not quite as stark a difference as in person, but it’s still noticeable.

Audio sample from the old fan

Audio sample from the new fan

And if you couldn’t hear it, here’s a visual representation:

Side-by-side comparison of the audio clips

The whole upgrade took about an hour (less time than writing this post about it!), and I even managed to avoid killing myself.

Opening the case and sticking my ear near to various components. High-tech stuff. ↩︎
On closer inspection it’s actually a slightly different model — there’s an S not a V in the middle there — but I’m going to assume that doesn’t change much on the quality/noise front. ↩︎
It then keeps spinning down to about 5V or so, it just needs enough power to initially get started. ↩︎

Building a new Computer

2025-06-03T00:00:00Z

The finished PC. Cat pen for scale.

I recently¹ built a new computer, after exclusively using a laptop for three years. It’s also the first time I’ve departed from the usual combo of an Intel CPU and Nvidia GPU.

While the form factor of a laptop did make it amazingly handy for travelling and attending LAN events, it was starting to show its age and there was basically no sane upgrade path. The main problem was its 3060 mobile graphics card, which was okay for the first few years and then slowly descended into painful. At the time, the (fairly disappointing) 50-series had just been released, but hadn’t yet made it into laptops. Nvidia had stopped production on the 40-series beforehand, so there also weren’t any compelling options there, either.

That’s not to say there were no laptops at all that I could have upgraded to. There were. Just not really any that ticked all of my perfectly normal boxes like “run Linux”, “have at least 64GB of ram”, and “run modern games well”. Having to upgrade the entire system just because the graphics card was showing its age was a bit of a drag, too. So back to the land of desktop computers I went!

To try to preserve some of that convenience, I opted for a small form factor (SFF) case. The computer and all the accessories can fit inside a hand-luggage-sized flight case. More on that later, though.

Components, Choices, and Cramming

So what, exactly, is in this computer? The case is a Fractal Design Terra, which is a lovely little 10.4L case². My last desktop PC was a normal-sized Fractal Design case, which I liked a lot, and it seems like they’ve upped their game since then. The Terra is both pleasant to look at (just look at that wooden panel!), and a breeze to work in. The sides and top are fully and easily removable, giving you great access to everything inside³. I expected an SFF build to be fiddly, but the case made it feel about the same as a normal full-sized build⁴.

The case with sides and top off, and wires everywhere in desperate need of management.

Another nice feature of the case is the movable spine. I think this is relatively common in SFF cases, but it was new to me. You mount the graphics card on one side of a metal divider, and everything else on the other. The divider (or spine) can be moved side-to-side depending on what exactly you’re putting in the case.

As I mentioned earlier, this was my first departure from the Intel + Nvidia bandwagon. Nvidia’s 50 series was underwhelming, and Intel managed to put out a generation of chips that literally fried themselves. It seemed like a good time to try out AMD’s offering! I went for the newly released Radeon RX 9070XT graphics card, and a Ryzen 7 9800X3D processor⁵. The processor has a Thermalright AXP90 X47 Full heatsink on top of it, with a Noctua NF-A9 fan on top of that.

The motherboard is a Gigabyte B850I Aorus Pro, and it hosts 2 32GB Corsair Vengeance RAM modules, and a 2TB Crucial T705 NVMe drive. Powering it all is a Cooler Master V750 SFX. The PSU is modular to reduce the number of wires that need to be fitted in, but they’re still pretty big and bulky. I’m tempted by custom cables but they’re rather pricey, and you can’t actually see inside with the sides closed…

With the diminutive size of the case and all the components crammed in, there’s actually no room left over for case fans. The PSU, GPU and CPU all have fans attached, but that’s it. There are stories online of people having heat issues, but thankfully everything seems pretty happy in my setup. The components all draw in air from the sides, and the hot air passively vents out the top. If I keep the CPU under maximum load for a while then it starts to thermal throttle slightly, but the only time I’ve actually managed that is when compiling a kernel. Why was I compiling a kernel, you ask? Well, we’ll come to that.

But does it Linux?

I didn’t go into this build completely blind. I did a bit of reading and chatting, and it sounded like AMD drivers were pretty good on Linux these days. I didn’t check specifically if any of the components were supported.

The 9070XT was released on the 6th of March. I built my PC on the 26th of March. The first version of Mesa (the 3D graphics lib) that was generally stable with 9070XT cards landed in Arch on the 20th of March. If I did the build just one week earlier, I would’ve had a much less fun experience.

It still wasn’t quite perfectly smooth, though. When playing games I’d occasionally have issues. In some cases the screen just went blank and came back after switching windows, but in others the graphics drivers couldn’t recover and the PC needed rebooting. Fortunately the drivers are open source, and they have a public issue tracker hosted by the Freedesktop project. Searching through the issues, I found they’d already been reported (by another Arch user, of course), and there were various suggestions for work-arounds, as well as lots of work trying to reproduce the issue.

Within a week, an AMD developer had posted a set of patches that might address the issues, and asked people to test them. That’s fortunately pretty easy on Arch: you just clone the repository containing the PKGBUILD file, drop in the patches, and update the checksums for them. Then you start compiling, realise you didn’t tell it to use more than one CPU core, abort it, and start again but this time running on all cylinders. Soon enough I was running a patched 6.14 kernel and the problems stopped completely. Others reported similar outcomes, and the patches worked their way into the kernel proper.

I spent a few weeks having to compile a custom kernel every time I accidentally updated it (I should’ve just stopped pacman from touching the kernel for a bit…), then one day the patches didn’t apply because they’d been merged upstream. Since then everything has been perfect.

A friend commented that “this wouldn’t happen with Nvidia”. They meant it negatively, but I wholeheartedly agree! Maybe Nvidia have fewer launch bugs than AMD, but you absolutely wouldn’t be able to talk to their driver engineers on a public bug tracker, and get access to test patches that fix your issue. I’ll happily take a few bugs in exchange for a company that operates openly in a way consistent with the rest of the Linux community.

But does it travel?

Travel mode, engage!

So the computer works fine at home, but how do I get it to other places? I just shoved a laptop in a backpack⁶, but you can’t quite do that with a PC even one this size. I’d seen a few people at LANs who packed everything in flight cases, but they always seemed a bit unwieldy. Doing some research, though, I found that I’d just be able to squeeze everything into a flight case sized to be taken onto a plane as cabin baggage.

Peli (or Pelican in the US) are the most well known maker of flight cases, but I chose to go with Nanuk instead. Spec-wise they’re fairly similar. Nanuk are a bit cheaper. The killer difference for me, though, is that Nanuk publish STEP files of their cases. As someone with a 3D printer, giving me CAD models of things so that I can easily mod them is definitely a way to win my favour.

After lots of measuring and thinking and modelling I came up with an arrangement where everything could fit snugly into the case. The case has cutouts for the wheels and handles, which constrain what can go where. I ended up wanting the computer at the handle-end of the case to make best use of space, but that means it needs something to take its weight when the case is vertical. The solution was, of course, 3D printing.

Using the STEP files from Nanuk I designed an insert that spans across the case, with screw holes that line up with the ones in the flight case. I wasn’t sure it would be able to take the weight like that, so I also added in some supports in the other direction (running down to the bottom of the case, if it’s stood up). I ended up printing it in three parts, and it came out perfectly first time. I can’t emphasise how nice it is having CAD models instead of having to measure and guess at things.

With the support in place, I focused on cushioning the computer. I opted for 10mm thick EVA foam as it’s easy to obtain and work with, and doesn’t take up too much room. I used contact adhesive to glue the foam directly to the case and the 3D printed divider. There’s a piece that goes between the computer and the keyboard, so I printed some little 3D supports to hold that more firmly in place, and stuck those in with contact adhesive as well.

So how does it all hold up? It’s been to one LAN event so far, and it’s great. I think the flight case might actually be easier to carry than hefting a heavy backpack: the handles are sturdy and comfortable, and it has chunky wheels for pulling it around. It’s also actually easier to get everything set up when you can see it all laid out in the flight case and pull out what you need, and then the case can be tucked away out the way for the duration of the event.

Recently-ish. It may or may not have taken me a few months to get around to writing this post. ↩︎
Using litres to measure computer cases always feels wrong. Those are the units we use for water. Don’t put them near the computers! ↩︎
The case also has holes to put security screws in to stop all of these nice easily removable parts from being removable. Very handy when you’re leaving it unattended at a LAN. ↩︎
Which for me means it was good fun for about an hour, and then I got fed up and just wanted it to be done already. ↩︎
Three months in and I still can’t remember these random strings of numbers. Someone even explained the versioning scheme to me. ↩︎
Along with a portable monitor, mouse, keyboard, headset, mouse mat, kensington lock, power supply, cables, and controller. It was a heavy bag. ↩︎

Coming around on LLMs

2025-05-28T00:00:00Z

Claude says hi.

For a long time I’ve been a sceptic of LLMs and how they’re being used and marketed. I tried ChatGPT when it first launched, and was totally underwhelmed. Don’t get me wrong: I find the technology damn impressive, but I just couldn’t see any use for it.

Recently I’ve seen more and more comments along the lines of “people who criticise LLMs haven’t used the latest models”, and a good number of developers that I respect have said they use coding models in some capacity. So it seemed like it was time to give them another shake.

The first decision to make was which model to try. OpenAI are no longer the only player in the game, every tech company of a certain size is now also somehow an AI company¹. I looked a bit at some benchmarks, and then mostly ignored them and went with the only company that I didn’t outright hate: Anthropic, and their model Claude.

Initial impressions

The latest Claude models do feel a lot more “capable” than the earlier ChatGPT versions I remember, but they also still have a lot of the same problems. At their heart, they’re still text-prediction models, and still seem to be trained to predict text that will please the user rather than be factually accurate or useful.

One of the improvements that people kept mentioning was the ability for models to access the web directly. I’ve seen it kick in a bit naturally, and asked for it explicitly sometimes, and it’s… nothing special? Are people just really bad at searching the web? Maybe they should try Kagi? I feel like about 80% of the time I could have found the information just as quickly myself, about 10% of the time it searched and then hallucinated an answer, and the remaining 10% it found something quicker than I otherwise would have. Those aren’t great results, especially when you consider how much cheaper and simpler just searching the web is.

There is another way to trigger web searches in Claude: using “research mode”. When you enable it, it splits off into different models: a “lead researcher” to come up with a plan, and then some minions that execute it. It ends up doing hundreds of web queries in the span of a few seconds. Suddenly I understand why things like Anubis need to exist. I’ve not used the “please launch a DoS attack” button since.

What did impress me, though, was its ability to churn out reasonable-ish code. It can hack together a bash script as well as I can, and do it far faster than I’d be able to. Sometimes they even work. That made me wonder what it would be like doing actual coding with it. Anthropic have a CLI tool called claude-code, so I paid them lots of money and gave it a spin.

Coding with Claude

The first thing I notice about claude-code is that I really like the interface. It’s basically an input box in a terminal. It doesn’t force me to use a certain IDE or do things in a certain way. By default it asks before making any changes, showing you a side-by-side diff of what it’s doing and allowing you to provide feedback. If I had to design a way to interact with a coding agent from scratch, I can’t think of many things I’d improve.

A simple example of a claude-code session

The power in claude-code versus just using the web UI is that it can use tools. It can query git, run grep commands, even use sed if it wants to change something in lots of files at once. It’s very good at figuring out its way around a codebase, even without any explicit instructions. You can see in the screenshot that with a little prompting it managed to get the permalink to this post; I didn’t tell it where the posts were stored, or how to work out the latest, I just told it when it was wrong. If I’d run /init before it would have probably picked up on the fact that my posts have custom permalinks, and noted it in CLAUDE.md.

But how good is it at actually writing code? It’s like having a keen but not particularly thorough Junior Engineer at your beck and call. If you give it a clearly defined task and guidance on how to implement it (and maybe some feedback as it suggests changes), it’s more than capable of doing it. If you don’t give it enough guidance it tends to go more off the rails. I tried having it generate a simple application from scratch with minimal technical guidance and no review of what it was doing, and it made such a mess of it I decided it was quicker to throw it away and start again by hand.

The man behind the curtain

Even with sufficient guidance, at times it’s really obvious that it’s an LLM generating pleasing-token-strings and not something that genuinely understands what it’s doing. It will spit out code like this:

if err != nil {
    if err == sql.ErrNoRows {
       return nil, err
    }
    return nil, err
}

Why’s that check for sql.ErrNoRows there? It’s entirely pointless. There was no instruction to check for it, none of the existing code checked for it, but I assume it comes up quite a bit in the training data. But it didn’t understand why, so it put the check in, and returned the exact same thing as if it hadn’t.

It also occasionally tries to “cheat” or solve problems the wrong way. It sometimes feels a bit like you’re asking for wishes from a Monkey Paw. “Stop the unit tests failing”, you’ll say; Claude will respond with a request to delete the failing test. The man behind the curtain isn’t particularly well hidden, and the training data and pattern matching often shows through.

This lack of understanding also makes it very hard to get Claude to use comments in a sensible way. It absolutely loves doing nonsense like this:

// render form
function renderForm() {
    // ...
}

Even with explicit instructions to avoid useless comments, or comments that only explain what the code does (not why it does it), or other prompts. Again, I’m chalking it up to a mixture of training data that does that, and lack of any actual understanding about why a human developer may want a comment to exist. Of course, there are plenty of flesh-and-blood devs out there that also don’t comment effectively, so maybe I shouldn’t give Claude too much of a hard time on this.

… and yet …

So with all those problems, it sounds like it’s just not worth it, right? Well, not quite. I feel like programming mostly consists of two distinct tasks: thinking how to implement things; actually implementing them; and then debugging why you’re off-by-one somewhere. Letting an LLM do the thinking is a big no-go for me: it’s evidently not good at it, and frankly that’s one of the things I most like about programming. Having let it loose on small projects, I shudder to think what the codebases of all the “vibe coded” projects that are popping up are like.

For the second stage, though, it’s actually quite nice. If I’ve thought through how I want something to be implemented, I can feed those steps to Claude and have it churn out the otherwise not-too-interesting code. This requires far less time and concentration on my part than writing code, to the extent that I can be thinking about the next feature, or doing something else at the same time. I don’t feel like I’m going to end up deskilling myself this way, as I’ve already formed the idea of what I want to code, I’m just using the LLM to spit it out faster than I can type it.

A lot of the most boring bits of coding like implementing CRUD-y operations can be summarised as “Look at this file/function. Do the same thing but slightly differently elsewhere”. Claude is great at this, especially when explicitly prompted like that. You’re basically playing to its pattern-matching strengths, rather than asking it to come up with anything novel. Most of my prompts tend to be prefixed with “Look at @some_file.go and @other_file.go.” to cue up the patterns I want it to use, and I find this works well.

As for debugging, it’s a mixed bag. It’s sometimes amazingly insightful, and sometimes just runs around in circles trying the wrong things over and over. It’s worth asking the question, but I definitely wouldn’t rely on it over my own abilities.

The future

I’m probably going to carry on using claude-code, at least for personal projects. I’ve got so much done that I just wouldn’t have been bothered to do if I was doing it all by hand. I very much enjoy being in the more “diffuse thinking” mindset, planning how things are going to work, rather than being stuck in the mines digging out SQL queries. After all, who wouldn’t want an over-eager assistant to work on all their hobby projects?

Work is a slightly different matter: a private project or even an open source project that disclaims any liability is different to something I’m being paid to deliver, and bear responsibility for fixing if it’s not done correctly. I’m not saying I won’t use it at all, but if I do it’ll be much more constrained than I would in personal projects.

As for non-code usages: I’m not sold. My sceptic hat is still firmly in place. I don’t think chat is a particularly good interface for many things, and hallucinations are still a big problem despite what people say. I hate the tide of AI slop that’s taking over the Internet, and how LLM-powered chat agents are being forced into every random product. You’re definitely not going to be seeing any AI-generated blog posts from me!

One topic I’ve not gone into here is the ethical concerns about using LLMs. They obviously exist, and I do have thoughts, but that’s a topic for another day.

I’m surprised it’s not gone more mainstream: why’s there no Tesco Value LLM model, yet? ↩︎

Adventures in 3D printing

2023-09-17T00:00:00Z

The Sovol SV06.

I’d been idly considering getting a 3D printer for a while, but have only recently taken the plunge. I picked up a Sovol SV06 from Amazon for £199.99¹, which is a model commonly recommended for beginners. About three weeks later, I think I’ve finally finished fixing all the problems the printer has, and thought I’d document them.

Setup and out of the box performance

The setup of the printer itself was straight forward. Most parts are assembled, you just have to bolt the frame together, bolt the various parts to the frame, and connect some wires. The hex bits in the iFixit Mako Bit Set were a godsend for this, as the included hex keys were a bit flimsy.

Once everything is bolted and connected, you turn it on and immediately start looking around for the jet engine you can hear. The fans are loud. The instructions walk you through the initial calibration, which consists of:

Levelling the X-Axis gantry (where the print head moves). The printer does this by (gently) ramming the gantry up as far as it will go.
Adjusting the Z-offset of the print head. The printer has an inductive proximity sensor attached to the extruder, but it doesn’t know how high it is relative to the nozzle. The calibration process involves manually dialing in this offset while moving a piece of paper back and forth under the nozzle.
Levelling the bed. This is entirely automatic: the printer just probes a bunch of points on the bed with the proximity sensor and builds up a mesh it can use to compensate for the bed not being level.

After that’s done, the included SD card has the standard “Benchy” model on it, pre-sliced and ready to print. It came out looking good:

A nice, problem free Benchy. No dramatic irony here.

The noise. Oh god, the noise.

I set about printing a few more things, but the noise of the machine quickly got to me. It sits approximately 80cm from my ear, and between the fans and the noises it makes when moving it was driving me mad. The fan in the PSU seems to sit at 100% power regardless of temperature or load, and sounds about three times louder than my laptop’s fans do when they’re running at full tilt. Fortunately, someone had designed a printable PSU fan silencer that redirects the airflow (and most of the noise) from the fan. It doesn’t feel like it should help that much, but it does significantly reduce the noise as it claims. This was the first of many times I’d use the 3D printer to print parts for itself.

The other noise was a lot less straightforward. The linear rods that all three axes run on pass through several bearings. Sovol apparently don’t bother to apply any lubrication to these, so it sounds like an elephant is moving through a gravel pit when the printer is running. Some people online suggest rubbing lube on the rods and pushing the movable parts back and forth, but this just results in a load of lube caught on the dust filters of the bearings².

The solution to this, then, was to dismantle most of the printer to gain access to each of the ten bearings, and manually lubricate them. This was tedious and messy, but not actually that difficult. It doesn’t seem like the kind of thing you should have to do on a brand-new machine, but I guess you get what you pay for.

Speeeeeeed

3D printers are slow. The SV06 is not one of the faster ones. One simple way to increase the print speed is to change out the nozzle. The SV06 comes with a 0.4mm nozzle, which is pretty standard, but if you upgrade to a 0.6mm nozzle you can extrude more than twice the volume of plastic in the same amount of time³. The consensus among anonymous internet people is that if you’re using a modern slicer there isn’t much difference in print quality when doing this.

I dutifully ordered a new nozzle, and after it turned up set about installing it. This is mostly straight forward, aside from the complication that you’re meant to do it when the hot-end is hot. That makes sense: you want it be under the same thermal expansion it will be under during use, but the tool they provide to change the nozzle is all metal. So you have a relatively delicate operation that you have to do in a bit of a rush before the heat creeps all the way up to your hand.

With the new nozzle in place, the printer was noticeably faster. I didn’t notice any difference in print quality, either.

SD cards are not fun

Up until this point, my process for printing anything was to slice it on the computer, export the gcode file to an SD card, turn around and walk over to the printer, insert the SD card, and start the print from the LCD. I don’t particularly mind those steps, but the SD card slot on the Sovol is terrible. It doesn’t line up well to the hole in the case, doesn’t guide the card in well, and is all round a pain to use.⁴

To remove the sneaker-net element of this process, I ordered a Raspberry Pi and installed OctoPrint on it. OctoPrint provides a web interface to the printer (which it controls over USB), and most slicers can automatically export the gcode to an OctoPrint instance.

While OctoPrint itself is great, there was one small problem: if you connect anything to the USB port on the printer, it tries to draw 5V from it and “backpower” itself. It’s not meant to draw power from the USB port, it’s just not designed well. A lot of the cheaper 3D printers share this flaw, it seems. The fix here is to get a power blocker such as the one made by PortaPow that simply leaves the +5V line unconnected between the two ends. Presumably Sovol could fix this by adding a diode to the USB power line, but maybe that would push their costs too high?

After getting OctoPrint working, I printed a nice case for it to mount on the side of the printer. I slotted the Pi in, and when I powered it on again it didn’t boot. When I went to pick it up to debug the problem, I burnt my finger on the SD card⁵. Apparently if you break SD cards in a certain way, they can short out and get very hot. Fun! Maybe I shouldn’t have been using the no-brand card that came with the printer…

What do we do with a drunken bed?

I continued printing bits and bobs, but was starting to notice a new issue: if I was printing multiple copies of the same object, those on the right hand side of the bed often failed. OctoPrint has a nifty plugin that lets you cancel certain regions from the build, which saves it ruining everything, but only being able to print in certain places was annoying.

I ran through all the troubleshooting steps I could think of, including a foray into custom firmware with different bed levelling abilities, but nothing seemed to help. Another OctoPrint plugin later and I could see the problem:

This should probably be straight.

The orientation of the graphic is a bit confusing, but you can see how the right hand side of the bed is significantly higher than the left. There’s a total variance of just over 0.5mm, which is apparently enough to cause problems even with the auto bed levelling. I spoke to Sovol support, and eventually they suggested disassembling the print bed and measuring the aluminium spacers that hold the bed. Sure enough, they were all different heights:

You had one job…

There’s a 0.13mm variance in their heights, which doesn’t account for the 0.5mm seen in the bed mesh, but it’s a good starting point. Rather than go back and forth with Sovol support over the course of many days, I just ordered some nylock nuts and reinstalled the bed without spacers. The nuts hold the bed in place, while still allowing you to tighten or loosen the screws to adjust the height.

After five or six rounds of totally and utterly uneventful adjusting and re-leveling, I got to this point:

This is much more straight.

There’s only so much you can do when you only have five screws to adjust, but I’m pretty happy the variance is now under 0.15mm. That’s definitely in the “good enough” range that the bed leveling algorithm can compensate for.

Whoops

So, about that “totally and utterly uneventful adjusting” I did. To get at the screws that hold the bed in place, you have to remove the thin sheet that acts as the build surface. This attaches magnetically to the bed, so it’s easy to do, but I am lazy and decided to leave it off while I fiddled with the bed.

There’s a slight problem here, though: when it’s working out the bed mesh the printer uses the inductive proximity sensor. The inductive proximity sensor detects metal. The metal is in the removable sheet. The first time I tried to run the bed mesh, the printer rammed the extruder assembly into the bed at a decent speed and ended up at a rather nasty angle.

I checked over the extruder and found the heat break had bent:

This should also probably be straight.

I tried printing with it as-is, but the filament came out too thin and at an angle, which made the print go about as well as you’d expect. I ordered a replacement, installed it, and… nothing worked still.

At this point I was getting thoroughly fed up. There seemed to be one problem after another, and no good way to diagnose problems with the print not sticking (because there are a dozen or so different potential causes). After a lot of frustrated googling, I found a post on the Prusa forums that suggested cleaning the build sheet with dish soap. I’d cleaned it with IPA, but several people were saying that dish soap magically fixed issues that IPA couldn’t. I took the plate to the kitchen, gave it a wash, and the next print went down perfectly.

Growing a backbone

For a blissful period of about four days the printer was chugging along perfectly with no issues. Then it started making a weird noise. In a display of amazing restraint, I didn’t immediately throw the whole thing out the window. Looking at the printer, I could see that the cable running from the extruder assembly to the main board was sagging and rubbing on top of the print as the print head was moving around. I’m not sure what property of the cable was keeping it aloft before, but it apparently ran out after about three weeks.

Like most of the other problems with the SV06, this design flaw has been noted by others and worked around. I printed a cable chain and installed it⁶. Amusingly, while I was printing the chain, the cable dragged on the print and dislodged one of the pieces entirely before it was finished. Fortunately, it didn’t affect the neighbouring pieces, so I just had to run off a replacement for that one.

The difference is rather pronounced:

It feels like I own a 3D printer to print parts for my 3D printer…

If you can’t see it: in the first image the cable is sagging and actually ends up below the print bed. The individual links of the cable chain limit its freedom of movement, stopping it sagging.

The end?

Once again the printer is printing properly again. I’m sure there will be more problems, but hopefully I get at least a week or so before anything else goes wrong. Maybe I’ll print an “X days since the last SV06 problem” counter…

I think, if I had to do this all again, but with the knowledge I have now, I’d have stumped up the extra cash to get a Prusa MK3S+⁷. This is the original machine on which the SV06 (and a lot of other cheap imitations) is based. Some obvious benefits it has over the SV06:

The extruder cable runs on the outside of the machine, so won’t droop onto prints
The heated bed has copper embedded in it for use in the induction sensor, so it won’t crash into the bed if the build sheet is removed
They actually put lubricant in the bearings instead of shipping them bone dry
The bed is held in place with screws instead of poorly sized aluminium spacers, so you can adjust the height without having to take the whole thing apart
They don’t use the cheapest possible power supply and main board, so it doesn’t backpower off of USB devices or have the loudest fans known to mankind

Prusa also have a much better attitude towards 3D printing: they open source the hardware designs (which is why there are so many cheap clones), and they actually use the printers themselves. The plastic parts of an SV06 are injection moulded, but those on a Prusa printer are actually printed on a Prusa printer. Not only is it cool that they use their own products, it means they have to run them en-masse so they’re obviously invested in making them reliable. They also have a reputation for exceptional support.

Despite all the problems, I do enjoy having the printer. I’ll probably talk more about the prints I’ve done that weren’t for the 3D printer itself in the future.

The RRP is £279.99, Amazon discounted it down to £269.99 then offered a £70 voucher. The voucher says it expires at midnight, but it’s said that every day for the past three weeks… ↩︎
Which is exactly what another set of people said would happen. It was worth a try, though. ↩︎
This feels strange, but if you work out the areas of the nozzles it makes sense: a 0.4mm diameter nozzle has an area of 0.04πmm² (the radius is half the diameter, and the area of a circle is πr²), and a 0.6mm diameter nozzle has an area of 0.09πmm². Maths! ↩︎
At least it doesn’t eject the SD card across the room like this Ender 3 V2! ↩︎
My only 3D printer related injury so far. It’s impressive I can be around a 200 degree hot-end and somehow burn myself on a storage device. ↩︎
Well actually I printed 5 copies of their calibration piece until it was perfect, then printed the actual cable chain, then had to sand each of the individual links because they still weren’t good enough. God, I hate sanding. ↩︎
Or maybe I’d cave and get the newer, shinier, much more expensive MK4. That wouldn’t be out of character. No, I didn’t just check my bank balance. You’re imagining things. ↩︎