Susam's Web Posts

I Will Not Add Query Strings to Your URLs

2026-05-09T00:00:00Z

Last evening, a short blog post appeared in my feed reader that felt as if it spoke directly to me. It is Chris Morgan's excellent post I've banned query strings.

Wisdom on the Web
Wander on the Web
Misfeature
Broken URLs
Qualms
Conclusion

Wisdom on the Web

Chris is someone whose Internet comments I have been reading for about half a decade now. I first stumbled upon his comments on Hacker News, where he left very detailed feedback on a small collection of boilerplate CSS rules I had shared there. I am by no means a web developer. I have spent most of my professional life doing systems programming in C and C++. However, developing websites and writing small HTML tools has been a long-time hobby for me. I have learnt most of my web development skills as a hobbyist by studying what other people do: first by viewing the source of websites I liked in the early 2000s and later by occasionally getting possessed by the urge to implement a new game or tool and searching MDN Web Docs to learn whatever I needed to make it work. One problem with learning a skill this way is that you sometimes pick up habits and practices that are fashionable but not necessarily optimal or correct. So it was really valuable to me when Chris commented on my collection of boilerplate CSS rules. It helped me improve my CSS a lot. In fact, a few of the lessons from his comment have really stuck with me; I keep them in mind whenever I make a hobby HTML project: always retain underlines in links and retain purple for visited links.

I have been following Chris's posts and comments on web-related topics since then. He often posts great feedback on web-related projects. Whenever I come across one, I make sure to read them carefully, even when the project isn't mine. I always end up learning something nice and useful from his comments. Here is one such recent example from the Lobsters story Adding author context to RSS.

Wander on the Web

A couple of months ago, I created a new project called Wander Console. It is a small, decentralised, self-hosted web console that lets visitors to your website explore interesting websites and pages recommended by a community of independent personal website owners. For example, my console is here: susam.net/wander/. If you click the 'Wander' button there, the tool loads a random personal web page recommended by the Wander community.

The tool consists of one HTML file that implements the console and one JavaScript file where the website owner defines a list of neighbouring consoles along with a list of web pages they recommend. If you copy these two files to your web server, you instantly have a Wander console live on the Web. You don't need any server-side logic or server-side software beyond a basic web server to run Wander Console. You can even host it in constrained environments like Codeberg Pages or GitHub Pages. When you click the 'Wander' button, the console connects to other remote consoles, fetches web page recommendations, picks one randomly and loads it in your web browser. It is a bit like the now defunct StumbleUpon but it is completely decentralised. It is also a bit like web rings except that the community network is not restricted to being a cycle; it is a graph that can take any shape.

There are currently over 50 websites hosting this tool. Together, they recommend over 1500 web pages. You can find a recent snapshot of the list of known consoles and the pages they recommend at susam.codeberg.page/wcn/. To learn more about this tool or to set it up on your website, please see codeberg.org/susam/wander.

Misfeature

In case you were wondering why I suddenly plugged my project into this post in the previous section, it is because I recently added a dubious feature to that project that I myself was not entirely convinced about. That misfeature is relevant to this post.

In version 0.4.0 of Wander Console, I added support for a via= query parameter while loading web pages. For example, if you encountered midnight.pub while using the console at susam.net/wander/, the console loaded the page using the following URL:

https://midnight.pub/?via=https://susam.net/wander/

This allowed the owner of the recommended website to see, via their access logs, that the visit originated from a Wander Console. Chris's recent blog post is critical of features like this. He writes:

I don't like people adding tracking stuff to URLs. Still less do I like people adding tracking stuff to my URLs.

https://chrismorgan.info/no-query-strings?ref=example.com? Did I ask? If I wanted to know I'd look at the Referer header; and if it isn't there, it's probably for a good reason. You abuse your users by adding that to the link.

I mentioned earlier that I was not entirely convinced that adding a referral query string was a good thing to do. Why did I add it anyway? I succumbed to popular demand. Let me briefly describe my frame of mind when I considered and implemented that feature. When I first saw the feature request on Codeberg, my initial reaction was reluctance. I wasn't convinced it was a good feature. But I was too busy with some ongoing algebraic graph theory research, another recent hobby, with a looming deadline, so I didn't have a lot of time to think about it clearly. In fact, everything about Wander Console has been made in very little time during the short breaks I used to take from my research. I made the first version of the console in about one and a half hours one early morning when my brain was too tired to read more algebraic graph theory literature and I really needed a break. During another such break, I revisited that feature request and, despite my reservations, decided to implement it anyway. During yet another such break, I am writing this post.

Normally, I don't like adding too many new features to my little projects. I want them to have a limited scope. I also want them to become stable over time. After a project has fulfilled some essential requirements I had, I just want to call it feature complete and never add another feature to it again. I'll fix bugs, of course. But I don't like to keep adding new features endlessly. That's my style of maintaining my hobby projects. So it should have been very easy for me to ignore the feature request for adding a referral query string to URLs loaded by the console tool. But I think a tired body and mind, worn down by long and intense research work, took a toll on me.

Although my gut feeling was telling me that it was not a good feature, I couldn't articulate to myself exactly why. So I implemented the referral query string feature anyway. While doing so, I added an opt-out mechanism to the configuration, so that if someone else didn't like the feature, they could disable it for themselves. This was another mistake. A questionable feature like this should be implemented as an opt-in feature, not an opt-out feature, if implemented at all. The fact that I didn't have a lot of time to reason through the implications of this feature meant that I just went ahead and implemented it without thinking about it critically. As the famous quote from Jurassic Park goes:

Your scientists were so preoccupied with whether or not they could that they didn't stop to think if they should.

Broken URLs

It soon turned out that my gut feeling was correct. After I implemented that feature, a page from one of my favourite websites refused to load in the console. To illustrate the problem, here are a few similar but slightly different URLs for that page:

The first and second URLs load fine, but the third URL returns an HTTP 404 error page. The website uses the query string to determine which one of its several font collections to show. So when we add an arbitrary query string to the URL, the website tries to interpret it as a font collection identifier and the page fails to load. That is why, when my tool added the via= query parameter to the first URL, the page failed to load.

Later, with a little time to breathe and some hindsight, I could articulate why adding referral query strings to a working URL was such a bad idea. Altering a URL gives you a new URL. The new URL could point to a completely different resource, or to no resource at all, even if the alteration is as small as adding a seemingly harmless query string. By adding the referral query string, I had effectively broken a working URL from a website I am very fond of.

Qualms

It is also worth asking whether an HTML tool should concern itself with referral query strings at all when web browsers already have a mechanism for this: the HTTP Referer header, governed by Referrer-Policy. That policy can be set at the server level, the document level or even on individual links. The Web standards already provide deliberate controls to decide how much referrer information should be sent. Appending referral query strings to URLs bypasses those controls. It moves a privacy and attribution concern out of the referrer mechanism and embeds it into the destination URL instead. I don't think an HTML tool should do that.

There is also a moral question here about whether it is okay to modify a given URL on behalf of the user in order to insert a referral query string into it. I think it isn't.

Conclusion

In the end, I decided to remove the referral query string feature from Wander Console. One might wonder why I couldn't simply leave the feature in as an opt-in. Well, the answer is that once I had deemed the feature misguided, I no longer wanted it to be part of my software in any form. The project is still new and we are still in the days of 0.x releases, so if there is a good time to remove features, this is it. But my ongoing research work left me with no time to do it. Finally, when the post I've banned query strings appeared in my feed reader last evening, it nudged me just enough to take a little time away from my academic hobby and devote it to removing that ill-considered feature. The feature is now gone. See commit b26d77c for details. The latest release, version 0.6.0, does not have it anymore.

This is a lesson I'll remember for any new hobby projects I happen to make in the future. If I ever load URLs again, I'll load them exactly as the website's author intended. I will never add query strings to your URLs.

Created using Simpsons Chalkboard Generator and GIMP

Read on website | #web | #technology

Wander Console 0.6.0

2026-05-08T00:00:00Z

Wander Console 0.6.0 is now available. This is the sixth release of Wander, a small, decentralised, self-hosted web console that lets visitors to your website discover interesting websites and pages recommended by a community of independent website owners. To try it, go to susam.net/wander/. To learn how it works and how to set it up on your own website, see the project README.

The main change in this release is the removal of support for the via referral query parameter, which was added to recommended URLs so that website owners could identify visits coming from a Wander Console in their access logs. The via query parameter was introduced in version 0.4.0 in response to community demand, but it turned out to be a misfeature. Some websites refuse to serve web pages when arbitrary query parameters are added to URLs. Although it was possible to turn this feature off via configuration, I would rather not keep a dubious feature that may prevent some pages from loading successfully. This feature has now been completely removed. See this excellent article by Chris Morgan to learn more about why adding arbitrary query parameters to URLs is a bad idea. Commit b26d77c has more details about the removal of this feature.

Apart from this change, there are a few minor user interface adjustments and fixes. See the changelog for more details.

If you own a personal website but have not set up a Wander Console yet, I suggest that you consider setting one up. You can see what it looks like by visiting mine at /wander/. To set up your own, follow the installation instructions in the README. It only involves copying two files to your web server, so the installation is very straightforward.

Also, check out our new community page called Wander Console Network. This page shows a recent snapshot of all known Wander Console instances along with the web pages they recommend. We have an IRC channel too at #wander on irc.libera.chat in case you are looking for a place to hang out with the community.

Update on 09 May 2026: See the post I Will Not Add Query Strings to Your URLs for an account of how the misguided referral query string feature was added and how it was removed.

Read on website | #web | #technology

From RSS to Atom

2026-05-04T00:00:00Z

Yesterday, I switched my website from RSS feeds to Atom feeds. In case you are wondering whether you have somehow landed on an ancient post from 2010, no, you have not. Yes, this is the year 2026, and I have finally switched from RSS feeds to Atom feeds. Yes, I am fifteen, or perhaps twenty, years too late.

Impulse Coding
Atom Entries
Temporary Workaround
Does It Matter?
References

Impulse Coding

I have always wanted to do this but could never make the time for it. Finally, it happened while I was giving my brain some rest from my ongoing algebraic graph theory studies. That's when I felt like spending a little time on my website and doing a little Lisp to change the feeds from RSS to Atom. I suppose this was impulse coding, a bit like impulse buying, except that I ended up with an Atom feed instead of a new book.

I find it quite surprising that when I have plenty of time, it usually does not occur to me to do these things, but when I am too busy and really short of time, these little ideas possess me during the short breaks I take. My personal website is one of my passion projects. Common Lisp is one of my favourite programming languages. So any time spent on this passion project using my favourite programming language is a very relaxing experience for me. It serves as an ideal break between intense study sessions. It took about an hour to implement the changes needed to make the switch from RSS to Atom. In the end, I could go back to my studies reinvigorated.

In case you are curious, here is the Git commit where I implemented the change from RSS to Atom: 596e1dd. As you might notice, a large portion of the change consists of replacing the key attribute in each post with the uuid attribute. The key attribute value was used as the value of the <guid> element in the RSS feeds. While an arbitrary short string could serve as the <guid> element for the items in an RSS feed, the <id> element of the entries in an Atom feed needs to be a URI. It turns out UUID URNs are a common choice for such a URI. I wrote the following shell command to replace all occurrences of the key attribute with uuid:

find . -type f -name '*.html' -exec sh -c '
  for f do
    sed "s/^<!-- key: .* -->/<!-- uuid: $(uuidgen) -->/g" "$f" > tmp &&
    mv tmp "$f"
  done
' sh {} +

The rest of the changes went into the feed templates and the Common Lisp program that statically generates the feeds along with the website.

For examples of the resulting feeds, see feed.xml and absurd.xml. The first is the main website feed and the second is an example of a tag-specific feed. Yes, the aforementioned Common Lisp program generates a feed for each tag. As of today, the main feed at feed.xml contains only two entries even though this website has over 200 pages. I explain the reason later in Temporary Workaround.

Atom Entries

Here is an example Atom entry from my feeds:

<entry>
  <title>A4 Paper Stories</title>
  <link href="https://susam.net/a4-paper-stories.html"/>
  <id>urn:uuid:06e5304d-c242-481c-bf94-e23b019b0a36</id>
  <updated>2026-01-06T00:00:00Z</updated>
  <content type="html">
    &lt;p&gt;I sometimes resort to a rather common measuring ...
  </content>
</entry>

The ellipsis (...) denotes content I have omitted for the sake of brevity.

I like how each entry in the feed now has its own UUIDv4. I also like that timestamps in an Atom feed are in the date-time format specified in RFC 3339, which also happens to be a profile of ISO 8601. Further, I like that I can explicitly declare the content type to be HTML. Commonly used values for the content type attribute are text, html and xhtml. If it is html, the content should be escaped HTML. If it is xhtml, the content should be an XHTML <div> element containing valid XHTML. Explicit content type support is likely the biggest advantage of Atom over RSS. In comparison, RSS 2.0 does not specify any way to declare the content type. So feed readers have to inspect the content and guess what the content type might be.

Temporary Workaround

As I mentioned before, as of today, the main feed contains only two entries. That's because only new posts published since the migration to Atom are now included in the feed. This was done to avoid spamming subscribers. The Atom specification's requirement that each entry's ID must be a URI has caused the IDs of every entry to change. If I were to include the older posts from before the change in the feed, then those posts would appear as new unread items. Subscribers can find this quite annoying. In fact, I have received a few complaints about this in the past. So I was careful this time. I have a little one-liner workaround in my site generator to exclude posts published before this change from the feed.

That was the only workaround I had to implement. Fortunately, my feed file had a neutral name like feed.xml, rather than a format-specific name like rss.xml, so I could avoid a URL change and the subsequent overhead of setting up redirects.

Does It Matter?

Does any of this matter today? I think it does. Contrary to the recurring claim that RSS and Atom are dead, most of the traffic to my personal website still comes from web feeds, even in 2026. Every time I publish a new post, I can see a good number of visitors arriving from feed readers. From the referrer data in my web server logs (which is not completely reliable but still offers some insight), the three largest sources of traffic to my website are web feeds, newsletters and search engines, in that order.

On the topic of newsletters, I was surprised to discover just how many technology newsletters there are on the Web and how active their user bases are. Once in a while, a newsletter picks up one of my silly or quirky posts, which then brings a large number of visits from its followers.

Back to the topic of web feeds, there is indeed a decent user base around RSS and Atom feeds. A good number of visitors to my website arrive by clicking a feed entry that shows up in their feed reader. I know this with some confidence by looking at the referer (sic) headers of visits to my HTML pages and the subsequent browsing of the website, as opposed to the isolated and automated fetches of the XML feeds. So there must be a reasonably active base of users around web feeds. It is a bit like being part of an invisible social network that we know exists and that we can measure through indirect evidence.

References

I found these three resources useful while switching to Atom feeds:

Read on website | #web | #technology

QuickQWERTY 1.2.3

2026-05-03T00:00:00Z

QuickQWERTY 1.2.3 is now available. QuickQWERTY is a web-based touch typing tutor for QWERTY keyboards that runs directly in the browser.

This release includes two small bug fixes. In the previous release, QuickQWERTY source code management moved from GitHub to Codeberg. During that update, the licence link in the footer was updated incorrectly. The broken licence link has now been fixed.

Further, there was a minor bug that caused a redundant dialog box to appear while switching between 6-7 split and 5-6 split. Units 16 to 20 contain two links labeled '6-7 split' and '5-6 split' which allow you to select how you want to split the number keys between left and right hands. Clicking either of those links brings up a dialog that explains what the two splits mean and prompts you to confirm in order to make the switch. Say, the '6-7 split' was already the chosen split. Clicking the '6-7 split' label triggered the dialog box unnecessarily. A dialog was unnecessary in that case since if you were already on the '6-7 split', clicking the '6-7 split' label resulted in no switch. This unnecessary dialog has now been eliminated.

To try out QuickQWERTY, please visit quickqwerty.html. The source code of QuickQWERTY is available under the terms of the MIT licence at codeberg.org/susam/quickqwerty.

Read on website | #web | #programming

Apr '26 Notes

2026-04-30T00:00:00Z

This is my fourth set of monthly notes for this year where I write down interesting facts and ideas I have explored during my spare time. There were three things in particular that occupied my leisure time this month. First, I managed to learn the proof of Tutte's famous theorem that any \( s \)-arc-transitive finite cubic graph must satisfy \( s \le 5. \) I learnt the proof from Norman Biggs's book Algebraic Graph Theory. The original proof appears in Tutte's 1947 paper 'A family of cubical graphs' (DOI). Biggs's presentation differs considerably from Tutte's original argument and relies heavily on the properties of stabiliser sequences of arcs. I should say that Biggs's proof, while complete, is extremely condensed. The proof reads more like a high-level outline that moves rapidly from one main result to the next without sufficiently explaining the intermediate steps. As a result, it took considerable effort to work out the proofs of the intermediate results. Biggs presents the proof in roughly nine pages spread across two chapters. However, when I worked through it in full detail while ensuring that every step is justified, my notes eventually grew to around 18 pages of A4 paper. The proof is quite involved, so I have not included it in these notes. Perhaps someday, when I have more time, I will distil my handwritten notes and publish them here on my website.

That was the first thing I spent time on this month. The second was revisiting some elementary results in group theory concerning cosets. I have found cosets to be an extremely useful concept that plays a central role in many areas of mathematics, including coding theory, Galois theory, field extensions and graph theory. In fact, Biggs's proof of Tutte's theorem discussed above also relies substantially on the theory of cosets. Since these results are relatively elementary and easier to write up, they are included in this set of notes.

Apart from mathematics, I also spent part of my spare time improving my new web project named Wander Console. These notes include some updates about this tool.

Coset Results
Computing

Coset Results

This section presents some very elementary results about cosets, together with brief proofs. These results appear repeatedly across many areas of mathematics and I often find myself using them almost instinctively, without consciously thinking through the underlying arguments each time. But once in a while, I like to sit back and ponder about their proofs from first principles, reflect on why they work and appreciate their elegance. The subsections below collect some of these little proofs.

Subgroup

Definition. Let \( G \) be a group with operation \( \cdot. \) Then a subset \( H \subseteq G \) is called a subgroup of \( G \) if \( H \) itself forms a group under the same operation and we write \( H \le G. \)

Cosets

Definition. Let \( G \) be a group, \( H \le G \) and \( a \in G. \) The left coset of \( H \) by \( a \) is the set \[ aH = \{ ah : h \in H \}. \] Similarly, the right coset of \( H \) by \( a \) is \[ Ha = \{ ha : h \in H \}. \]

Coset Membership

Theorem. Let \( G \) be a group with identity \( e, \) \( H \le G \) and \( a \in G. \) Then \( aH = H \) if and only if \( a \in H. \)

Proof. Suppose \( aH = H. \) Then \[ aH = H \implies ae \in aH = H \implies a \in H. \] Conversely, suppose \( a \in H. \) Let \( x \in aH. \) Then \( x = ah \) for some \( h \in H. \) Then by the closure property of the subgroup \( H, \) we get \[ a, h \in H \implies ah \in H \implies x \in H. \] Thus \( aH \subseteq H. \) To show the reverse inclusion, let \( x \in H. \) Since \( a \in H, \) we have \( a^{-1} \in H, \) so \[ x \in H \implies a^{-1} x \in H \implies a(a^{-1} x) \in aH \implies x \in aH. \] Therefore \( H \subseteq aH. \) As a result, \( H = aH. \)

Coset Equality

Theorem. Let \( G \) be a group with identity \( e, \) \( H \le G \) and \( a, b \in G. \) Then \( aH = bH \) if and only if \( a^{-1} b \in H. \)

Proof. Suppose \( aH = bH. \) Then \begin{align*} aH = bH & \implies b \in bH = aH \\ & \implies b = ah \tag{for some \( h \in H \)}\\ & \implies a^{-1} b = h \\ & \implies a^{-1} b \in H. \end{align*} Conversely, suppose \( a^{-1} b \in H. \) Let \( h = a^{-1} b. \) Then \begin{align*} h = a^{-1} b & \implies ah = b \\ & \implies (ah)H = bH \\ & \implies a(hH) = bH \\ & \implies aH = bH \tag{since \( hH = H \)}. \end{align*}

Corollary. Let \( G \) be a group with identity \( e, \) \( H \le G \) and \( a, b \in G. \) Then \( aH = bH \) if and only if \( a \in bH. \)

Proof. \[ aH = bH \iff b^{-1} a \in H \iff b (b^{-1} a) \in bH \iff a \in bH. \]

Partitioning a Group into Disjoint Cosets

Theorem. Let \( G \) be a group with identity \( e, \) \( H \le G \) and \( a, b \in G. \) Then either \( aH \cap bH = \varnothing \) or \( aH = bH. \)

Proof. Suppose \( aH \cap bH \ne \varnothing. \) Then there exists some element \( x \in aH \cap bH. \) Then \[ x = ah_1 = bh_2 \] for some \( h_1, h_2 \in H. \) Therefore \[ a^{-1} b = h_1 h_2^{-1} \in H. \] Then by section Coset Equality, \[ aH = bH. \]

Computing

Wander Console Updates

Wander Console is a new project I developed in March while taking a short break from my algebraic graph theory studies. It is a tiny, decentralised, self-hosted web console that allows visitors to a website to explore other interesting personal websites. It is similar to the now-defunct service named StumbleUpon, but unlike StumbleUpon it has no central service and no server-side logic. Wander is hosted entirely on independent personal websites. Wander Consoles link to one another and fetch web page recommendations from each other. The entire tool consists of just two files: an HTML file and a JS file. Everything, including connecting to other Wander Consoles in the network and recommending webpages, happens entirely on the client side in the user's web browser. See codeberg.org/susam/wander for more details.

Two releases of Wander were made this month. One notable feature introduced this month is the console network crawler. You can visit any Wander console and use this feature to crawl the network reachable from it. To see this in action, go to my console at wander/, click the Console button at the top and then click the Crawl button.

Since different consoles link to different sets of peers, each one has its own neighbourhood, so the crawler output varies from console to console.

Due to the decentralised nature of the tool, it is difficult to know exactly how many people have set up Wander Console on their websites. Nevertheless, I used the crawler with a known set of consoles to explore as much of the network as possible. The result is available at susam.codeberg.page/wcn/. It currently shows over 50 consoles recommending more than 1400 web pages from the small web of personal sites. For a project that is only six weeks old, these seem like decent numbers.

Flush Output

When I log into my Debian 13.2 system via SSH, I find that ctrl+o does not enable output discarding.

$ stty -a | grep 'discard\|flush'
werase = ^W; lnext = ^V; discard = ^O; min = 1; time = 0;
echoctl echoke -flusho -extproc
$ ping 127.0.0.1
PING 127.0.0.1 (127.0.0.1) 56(84) bytes of data.
64 bytes from 127.0.0.1: icmp_seq=1 ttl=64 time=0.021 ms
64 bytes from 127.0.0.1: icmp_seq=2 ttl=64 time=0.030 ms
^O64 bytes from 127.0.0.1: icmp_seq=3 ttl=64 time=0.026 ms
64 bytes from 127.0.0.1: icmp_seq=4 ttl=64 time=0.049 ms
^C
--- 127.0.0.1 ping statistics ---
4 packets transmitted, 4 received, 0% packet loss, time 3079ms
rtt min/avg/max/mdev = 0.021/0.031/0.049/0.010 ms

On macOS 15.3.2, with the Terminal app, it does work as expected.

$ stty -a | grep 'discard\|flush'
	-echoprt -altwerase -noflsh -tostop -flusho pendin -nokerninfo
cchars: discard = ^O; dsusp = ^Y; eof = ^D; eol = <undef>
$ ping 127.0.0.1
PING 127.0.0.1 (127.0.0.1): 56 data bytes
64 bytes from 127.0.0.1: icmp_seq=0 ttl=64 time=0.087 ms
64 bytes from 127.0.0.1: icmp_seq=1 ttl=64 time=0.221 ms
^O^C
--- 127.0.0.1 ping statistics ---
9 packets transmitted, 9 packets received, 0.0% packet loss
round-trip min/avg/max/stddev = 0.068/0.171/0.362/0.087 ms

Other Stuff

My initial draft of this post had a few additional sections that have since been moved to their own posts:

Read on website | #notes | #mathematics | #web | #technology

QuickQWERTY 1.2.2

2026-04-28T00:00:00Z

QuickQWERTY 1.2.2 is now available. QuickQWERTY is a web-based touch typing tutor for QWERTY keyboards that runs directly in the browser.

This release includes two important changes. First, a longstanding bug in the practice pane has been fixed. While practising a lesson, a 'Restart' link appears in the practice pane. Due to a bug, clicking it incorrectly sent users to Unit 1.1 instead of restarting the current lesson. The link now correctly restarts the active lesson.

Second, source code hosting has moved to Codeberg. Codeberg is the third home of this 17 year old project. QuickQWERTY was first hosted on SourceForge in 2008, where it remained for seven years. In 2015, the project moved to GitHub. It has now moved once again, this time to Codeberg.

As before, QuickQWERTY continues to be available under the terms of the MIT licence. The latest version remains available on this website at quickqwerty.html.

Read on website | #web | #programming

Wander Console 0.5.0

2026-04-19T00:00:00Z

Wander Console 0.5.0 is out. It is the fifth release of Wander, a small, decentralised, self-hosted web console that lets visitors to your website explore interesting websites and pages recommended by a community of independent website owners. To try it, go to susam.net/wander/. To learn more about how it works and how to set it up on your website, see the project README.

A screenshot of Wander Console

The big feature in this release is a built-in console network crawler. To try the console crawler, go to susam.net/wander/ > Console > Crawl. It performs a breadth-first search (BFS) traversal of the Wander network and lists all discovered consoles and page recommendations in a single pane.

A screenshot of Wander Console Crawler

If you have set up a Wander Console instance for yourself on your website, I recommend upgrading to the latest version to use this feature. It is fun to find out just how many Wander consoles belong to your neighbourhood. To upgrade, you only need to download the Wander Console bundle mentioned here and replace your existing Wander index.html with the new one.

If you own a personal website but have not set up a Wander Console yet, I suggest that you consider setting one up for yourself. You can see what it looks like by visiting mine at /wander/. To set up your own, follow these instructions from the README: Install. It just involves copying two files to your web server. It is as simple as that.

Read on website | #web | #technology

Wander Console 0.4.0

2026-04-04T00:00:00Z

Wander Console 0.4.0 is the fourth release of Wander, a small, decentralised, self-hosted web console that lets visitors to your website explore interesting websites and pages recommended by a community of independent website owners. To try it, go to susam.net/wander/.

A screenshot of Wander Console 0.4.0

This release brings a few small additions as well as a few minor fixes. You can find the previous release pages here: /code/news/wander/. The sections below discuss the current release.

Wildcard Patterns
The 'via' Query Parameter
Console Picker Algorithm
Allow Links that Open in New Tab
Community

Wildcard Patterns

Wander Console now supports wildcard patterns in ignore lists. An asterisk (*) anywhere in an ignore pattern matches zero or more characters in URLs. For example, an ignore pattern like https://*.midreadpopup.example/ can be used to ignore URLs such as this:

https://alice.midreadpopup.example/
https://bob.jones.midreadpopup.example/

These ignore patterns are specified in a console's wander.js file. These are very important for providing a good wandering experience to visitors. The owner of a console decides what links they want to ignore in their ignore patterns. The ignore list typically contains commercial websites that do not fit the spirit of the small web, as well as defunct or incompatible websites that do not load in the console. A console with a well maintained ignore list ensures that a visitor to that console has a lower likelihood of encountering commercial or broken websites.

For a complete description of the ignore patterns, see Customise Ignore List.

The 'via' Query Parameter

By popular demand, Wander now adds a via= query parameter while loading a recommended web page in the console. The value of this parameter is the console that loaded the recommended page. For example, if you encounter midnight.pub/ while using the console at susam.net/wander/, the console loads the page using the following URL:

https://midnight.pub/?via=https://susam.net/wander/

This allows the owner of the recommended website to see, via their access logs, that the visit originated from a Wander Console. While this is the default behaviour now, it can be customised in two ways. The value can be changed from the full URL of the Wander Console to a small identifier that identifies the version of Wander Console used (e.g. via=wander-0.4.0). The query parameter can be disabled as well. For more details, see Customise 'via' Parameter.

Console Picker Algorithm

In earlier versions of the console, when a visitor came to your console to explore the Wander network, it picked the first recommendation from the list of recommended pages in it (i.e. your wander.js file). But subsequent recommendations came from your neighbours' consoles and then their neighbours' consoles and so on recursively. Your console (the starting console) was not considered again unless some other console in the network linked back to your console.

A common way to ensure that your console was also considered in subsequent recommendations too was to add a link to your console in your own console (i.e. in your wander.js). Yes, this created self-loops in the network but this wasn't considered a problem. In fact, this was considered desirable, so that when the console picked a console from the pool of discovered consoles to find the next recommendation, it considered itself to be part of the pool. This workaround is no longer necessary.

Since version 0.4.0 of Wander, each console will always consider itself to be part of the pool from which it picks consoles. This means that the web pages recommended by the starting console have a fair chance of being picked for the next web page recommendation.

Allow Links that Open in New Tab

The Wander Console loads the recommended web pages in an <iframe> element that has sandbox restrictions enabled. The sandbox properties restrict the side effects the loaded web page can have on the parent Wander Console window. For example, with the sandbox restrictions enabled, a loaded web page cannot redirect the parent window to another website. In fact, these days most modern browsers block this and show a warning anyway, but we also block this at a sandbox level too in the console implementation.

It turned out that our aggressive sandbox restrictions also blocked legitimate websites from opening a link in a new tab. We decided that opening a link in a new tab is harmless behaviour and we have relaxed the sandbox restrictions a little bit to allow it. Of course, when you click such a link within Wander console, the link will open in a new tab of your web browser (not within Wander Console, as the console does not have any notion of tabs).

Community

Although I developed this project on a whim, one early morning while taking a short break from my ongoing studies of algebraic graph theory, the subsequent warm reception on Hacker News and Lobsters has led to a growing community of Wander Console owners. There are two places where the community hangs out at the moment:

New consoles are announced in this thread on Codeberg: Share Your Wander Console.
We also have an Internet Relay Chat (IRC) channel named #wander on the Libera IRC network. This is a channel for people who enjoy building personal websites and want to talk to each other. You are welcome to join this channel, share your console URL, link to your website or recent articles as well as share links to other non-commercial personal websites.

If you own a personal website but you have not set up a Wander Console yet, I suggest that you consider setting one up for yourself. You can see what it looks like by visiting mine at /wander/. To set up your own, follow these instructions: Install. It just involves copying two files to your web server. It is about as simple as it gets.

Read on website | #web | #technology

Wander Console 0.3.0

2026-03-25T00:00:00Z

Wander 0.3.0 is the third release of Wander, a small, decentralised, self-hosted web console that lets visitors to your website explore interesting websites and pages recommended by a community of independent website owners. To try it, go to susam.net/wander/.

This release brings small but important bug fixes. The previous release, version 0.2.0 introduced a number of new features. Unfortunately, two of them caused issues for some users. A new feature in the previous release was the ignore list feature. The ignore list defines console URLs and page URLs that the console never uses while discovering page recommendations. While this feature works fine, due to a bug in the implementation, the Console dialog fails to load in consoles that do not define any ignore list. This has now been fixed.

There was another issue due to which the <iframe> that displays discovered websites and pages could not load certain websites. In particular, any website that relied on same-origin context to load its own resources failed to load in the console. This has been fixed as well. Please see codeberg.org/susam/wander/issues/7 for a detailed discussion on this issue.

Apart from these two important fixes, there are a few other minor fixes too pertaining to preventing horizontal scrolling in small devices and preventing duplicate recommendations from appearing too close to each other. Please see CHANGES.md for a detailed changelog.

To learn more about Wander, how it works and how to set it up, please read the project README at codeberg.org/susam/wander. To try it out right now, go to susam.net/wander/.

Read on website | #web | #technology

Wander Console 0.2.0

2026-03-24T00:00:00Z

Wander 0.2.0 is the second release of Wander, a small, decentralised, self-hosted web console that lets visitors to your website explore interesting websites and pages recommended by a community of independent personal website owners. To try it, go to susam.net/wander.

This release brings a number of improvements. When I released version 0.1.0, it was the initial version of the software I was using for my own website. Naturally, I was the only user initially and I only added trusted web pages to the recommendation list of my console. But ever since I announced this project on Hacker News, it has received a good amount of attention. It has been less than a week since I announced it there but over 30 people have set up a Wander console on their personal websites. There are now over a hundred web pages being recommended by this network of consoles. With the growth in the number of people who have set up Wander console, came several feature requests, most of which have been implemented already. This release makes these new features available.

Since Wander 0.2.0, the wander.js file of remote consoles is executed in a sandbox iframe to ensure that it has no side effects on the parent Wander console page. Similarly, the pages recommended by the network are also loaded into a sandbox iframe.

This release also brings several customisation features. Console owners can customise their Wander console by adding custom CSS or JavaScript. Console owners can also block certain URLs from ever being recommended on their console. This is especially important in providing a good wandering experience to visitors. Since this network is completely decentralised, console owners can add any web page they like to their console. Sometimes they inadvertently add pages that do not load successfully in the console due to frame embedding restrictions. This leads to an uneven wandering experience because these page recommendations occasionally make it to other consoles where they fail to load. Console owners can now block such URLs in their console to decrease the likelihood of these failed page loads. This helps make the wandering experience smoother.

Another significant feature in this release is the expanded Console dialog box. This dialog box now shows various details about the console and the current wandering session. For example, it shows the console's configuration: recommended pages, ignored URLs and linked consoles. It also shows a wandering history screen where you can see each link that was recommended to you along with the console that recommendation came from. There is another screen that shows all the consoles discovered during the discovery process. Those who care about how Wander works would find this dialog box quite useful. To check it out, go to my Wander console and explore.

To learn more about Wander, how it works and how to set it up, please read the project README at codeberg.org/susam/wander.

Read on website | #web | #technology

Wander Console 0.1.0

2026-03-18T00:00:00Z

Wander 0.1.0 is the first release of Wander, a small, decentralised, self-hosted web console that lets visitors to your website explore interesting websites and pages recommended by a community of independent personal website owners.

Anyone with a personal website can take this tool and host an instance of a Wander console. Each Wander console loads personal websites and pages recommended by the Wander community. Further, each Wander console can link to other Wander consoles, forming a lightweight, decentralised network for browsing the small web of personal websites.

Setting up an instance of a Wander console involves copying just two static files from the Wander project at codeberg.org/susam/wander. The most interesting aspect of the Wander console is that discovery of new links from other consoles happens on the client side in the user's web browser. As a website owner, you do not need to set up any server-side components beyond a basic web server. In fact, you can host a Wander console on GitHub Pages or Codeberg Pages too.

To learn more about Wander, how it works and how to set it up, please read the project README at codeberg.org/susam/wander.

Read on website | #web | #technology

Wander Console

2026-03-16T00:00:00Z

I have put together a small tool to explore the small web of personal websites. It is called Wander. Please visit susam.net/wander/ to try out my Wander console.

If you have your own website, please consider joining this community by hosting your own Wander console. To do so, visit codeberg.org/susam/wander and follow the instructions there. Thank you!

Read on website | #html | #web | #technology

HN Skins 0.4.0

2026-03-10T00:00:00Z

HN Skins 0.4.0 is a minor update to HN Skins, a web browser userscript that adds custom themes to Hacker News and lets you browse HN with a variety of visual styles. This release introduces a small fix to preserve the commemorative black bar that occasionally appears at the top of the page.

When a notable figure in technology or science passes away, Hacker News places a thin black bar at the top of the page in tribute. Previously some skins could obscure this element. This update ensures that the bar remains visible and clearly noticeable. In dark themed skins, the black bar is rendered as a lighter shade of grey so that it maintains sufficient contrast and remains conspicuous.

Today Hacker News has a story about Tony Hoare passing away, which made me notice that the commemorative black bar was not rendered properly with some skins. This prompted me to investigate the issue and implement the fix included in this release.

Screenshots showing how the bar appears with different skins are available at susam.github.io/blob/img/hnskins/0.4.0/.

To install HN Skins, visit github.com/susam/hnskins and follow the instructions there.

Read on website | #web | #programming | #technology

HN Skins 0.3.0

2026-03-07T00:00:00Z

HN Skins 0.3.0 is a minor update to HN Skins, a web browser userscript that adds custom themes to Hacker News and allows you to browse HN with a variety of visual styles. This release includes fixes for a few issues that slipped through earlier versions. For example, the comment input textbox now uses the same font face and size as the rest of the active theme. The colour of visited links has also been slightly muted to make it easier to distinguish them from unvisited links. In addition, some skins have been renamed: Teletype is now called Courier and Nox is now called Midnight.

Further, the font face of several monospace based themes is now set to monospace instead of courier. This allows the browser's preferred monospace font to be used. The font face of the Courier skin (formerly known as Teletype) remains set to courier. This will never change because the sole purpose of this skin is to celebrate this legendary font.

To view screenshots of HN Skins or install it, visit github.com/susam/hnskins.

Read on website | #web | #programming | #technology

HN Skins 0.2.0

2026-03-01T00:00:00Z

HN Skins 0.2.0 is a minor update of HN Skins. It comes a day after its initial release in order to fine tune a few minor issues with the styles in the initial release. HN Skins is a web browser userscript that adds custom themes to Hacker News and allows you to browse HN with different visual styles.

This update removes excessive vertical space below the 'reply' links, sorts the skin options alphabetically in the selection dialog and fixes the background colour of the navigation bar in the Terminal skin by changing it from a dark grey to a dark green.

Soon after making this release, I discovered a few other minor issues, such as the Cafe and Terminal themes using Courier when I intended them to use the system monospace font. This has already been fixed in the development version currently available on GitHub. However, I will make a formal release later.

See the changelog for more details. To see some screenshots of HN Skins or to install it, visit github.com/susam/hnlinks and follow the instructions there.

Read on website | #web | #programming | #technology

HN Skins 0.1.0

2026-02-28T00:00:00Z

HN Skins 0.1.0 is the initial release of HN Skins, a browser userscript that adds custom themes to Hacker News (HN). It allows you to browse HN in style with a selection of visual skins.

To use HN Skins, first install a userscript manager such as Greasemonkey, Tampermonkey or Violentmonkey in your web browser. Once installed, you can install HN Skins from github.com/susam/hnskins.

The source code is available under the terms of the MIT licence. For usage instructions and screenshots, please visit github.com/susam/hnskins.

Read on website | #web | #programming | #technology

Nerd Quiz #4

2026-02-22T00:00:00Z

Nerd Quiz #4 is the fourth instalment of Nerd Quiz, a single page HTML application that challenges you to measure your inner geek with a brief quiz. Each question in the quiz comes from everyday moments of reading, writing, thinking, learning and exploring.

This release introduces five new questions drawn from a range of topics, including computing history, graph theory and Unix. Visit Nerd Quiz to try the quiz.

A community discussion page is available here. You are very welcome to share your score or discuss the questions there.

Read on website | #web | #miscellaneous | #game

QuickQWERTY 1.2.1

2026-01-27T00:00:00Z

QuickQWERTY 1.2.1 is now available. QuickQWERTY is a web-based touch typing tutor for QWERTY keyboards that runs directly in the web browser.

This release contains a minor bug fix in Unit 4.3. Unit 4.3 is a 'Control' unit that lets you practise typing partial words as well as full words. In one place in this unit, the following sequence of partial and full words occurs:

l li lime lime

The full word lime was incorrectly repeated twice. This has been fixed to:

l li lim lime

To try out QuickQWERTY, go to quickqwerty.html.

Read on website | #web | #programming

Attention Media ≠ Social Networks

2026-01-20T00:00:00Z

When web-based social networks started flourishing nearly two decades ago, they were genuinely social networks. You would sign up for a popular service, follow people you knew or liked and read updates from them. When you posted something, your followers would receive your updates as well. Notifications were genuine. The little icons in the top bar would light up because someone had sent you a direct message or engaged with something you had posted. There was also, at the beginning of this millennium, a general sense of hope and optimism around technology, computers and the Internet. Social networking platforms were one of the services that were part of what was called Web 2.0, a term used for websites built around user participation and interaction. It felt as though the information superhighway was finally reaching its potential. But sometime between 2012 and 2016, things took a turn for the worse.

First came the infamous infinite scroll. I remember feeling uneasy the first time a web page no longer had a bottom. Logically, I knew very well that everything a browser displays is a virtual construct. There is no physical page. It is just pixels pretending to be one. Still, my brain had learned to treat web pages as objects with a beginning and an end. The sudden disappearance of that end disturbed my sense of ease.

Then came the bogus notifications. What had once been meaningful signals turned into arbitrary prompts. Someone you followed had posted something unremarkable and the platform would surface it as a notification anyway. It didn't matter whether the notification was relevant to me. The notification system stopped serving me and started serving itself. It felt like a violation of an unspoken agreement between users and services. Despite all that, these platforms still remained social in some diluted sense. Yes, the notifications were manipulative, but they were at least about people I actually knew or had chosen to follow. That, too, would change.

Over time, my timeline contained fewer and fewer posts from friends and more and more content from random strangers. Using these services began to feel like standing in front of a blaring loudspeaker, broadcasting fragments of conversations from all over the world directly in my face. That was when I gave up on these services. There was nothing social about them anymore. They had become attention media. My attention is precious to me. I cannot spend it mindlessly scrolling through videos that have neither relevance nor substance.

But where one avenue disappeared, another emerged. A few years ago, I stumbled upon Mastodon and it reminded me of the early days of Twitter. Back in 2006, I followed a small number of folks of the nerd variety on Twitter and received genuinely interesting updates from them. But when I log into the ruins of those older platforms now, all I see are random videos presented to me for reasons I can neither infer nor care about. Mastodon, by contrast, still feels like social networking in the original sense. I follow a small number of people I genuinely find interesting and I receive their updates and only their updates. What I see is the result of my own choices rather than a system trying to capture and monetise my attention. There are no bogus notifications. The timeline feels calm and predictable. If there are no new updates from people I follow, there is nothing to see. It feels closer to how social networks used to work originally. I hope it stays that way.

Read on website | #web | #technology

Nested Code Fences in Markdown

2026-01-19T00:00:00Z

Today, we will meet a spiky-haired nerd named Corey Dumm, who normally lives within Markdown code fences. We will get to know him a bit, smile with him when his fences hold and weep quietly when misfortune strikes.

One of the caveats of the Markdown universe is the wide variety of Markdown implementations available. In these parallel universes, the rules of Markdown rendering differ subtly. In this post, we will focus only on the CommonMark specification. Since GitHub Flavoured Markdown (GFM) is a strict superset of CommonMark, whatever we discuss here applies equally well to both CommonMark and GFM.

Basic Code Fences
Fancy Code Fences
Basic Code Spans
Fancy Code Spans
Specification

Basic Code Fences

Corey had a knack for working with computers ever since he was a kid.

Corey at his computer:

```
(o_o)--.|[_]|
```

Everything was perfect in Corey's world. The CommonMark renderer would convert the Markdown above to the following HTML:

Corey at his computer:

(o_o)--.|[_]|

View HTML

<p>Corey at his computer:</p>
<pre><code>(o_o)--.|[_]|
</code></pre>

At this point, all was well. Corey grew quickly. Before long, he had a head full of spiky hair. Then the fences began to matter.

Corey, all grown up:

```
 ```
(o_o)--.|[_]|
```

Let us see how this renders. I must warn you that during the Markdown-to-HTML translation, Corey loses his hair. Some viewers may find the following scene disturbing. Viewer discretion is advised. Here is the rendered HTML:

Corey, all grown up:

(o_o)--.|[_]|

View HTML

<p>Corey, all grown up:</p>
<pre><code></code></pre>
<p>(o_o)--.|[_]|</p>
<pre><code></code></pre>

Corey's hair is gone! What a catastrophic accident! Corey is alright, though. He is still quite afraid of Markdown fences, but otherwise well and bouncing back. Why did this happen? The second set of triple backticks immediately ends the fenced code block started by the first set of triple backticks. As a result, Corey's smiley face ends up outside the fenced code block. The triple backticks that were once Corey's hair are now woven into the fabric of the surrounding HTML. Fortunately, CommonMark offers a few ways to avoid such accidents.

Fancy Code Fences

In CommonMark, fenced code blocks are most commonly written using triple backticks. However, the specification also allows tildes as an alternative fence. This can be useful when the code itself contains backticks. Let us see an example:

Corey, all grown up:

~~~
 ```
(o_o)--.|[_]|
~~~

In fact, a code fence need not consist of exactly three backticks or tildes. Any number of backticks or tildes is allowed, as long as that number is at least three. The following is therefore equivalent:

Corey, all grown up:

~~~~~
 ```
(o_o)--.|[_]|
~~~~~

And so is this:

Corey, all grown up:

`````
 ```
(o_o)--.|[_]|
`````

All three examples render like this:

Corey, all grown up:

 ```
(o_o)--.|[_]|

View HTML

<p>Corey, all grown up:</p>
<pre><code> ```
(o_o)--.|[_]|
</code></pre>

No hair is lost in translation.

Basic Code Spans

A similar problem arises with inline code spans. Most Markdown users know to use backticks to delimit inline code spans. For example:

An old picture of Corey at his computer: `(o_o)--.|[_]|`

This produces the following output:

An old picture of Corey at his computer: (o_o)--.|[_]|

View HTML

<p>An old picture of Corey at his computer: <code>(o_o)--.|[_]|</code></p>

However, what do we do when we need to put Corey's dear friend Becky Trace within an inline code span? Becky has short, straight hair tucked neatly on either side of her face. Here's a picture of her:

`(o_o)`

I believe you can already see the difficulty here. Inline code spans use backticks as delimiters. So when we put Becky within a code span, the first backtick in Becky's face would terminate the code span immediately and then the rest of Becky would lie outside it. CommonMark offers solutions for this kind of situation as well.

Fancy Code Spans

An inline code span delimiter need not consist of exactly one backtick. It can consist of any number of backticks. So `foo` and ``foo`` produce identical HTML. There is another important but less well-known detail. When the text inside an inline code span begins and ends with spaces, one space is removed from each end before rendering. So `foo` and ` foo ` are equivalent. Therefore, when we need to put backticks within an inline code span, we can start the code span using multiple backticks and a space. For example:

Meet Corey's friend Becky Trace: `` `(o_o)` ``

Here is the rendered output:

Meet Corey's friend Becky Trace: `(o_o)`

View HTML

<p>Meet Corey's friend Becky Trace: <code>`(o_o)`</code></p>

Becky has her hair intact too. We have avoided the mishap that once caused great distress to Corey. That, my friends, is how backticks survive nesting in Markdown.

Specification

Before I finish this post, let us take a look at the CommonMark specification to see where these details are defined. The excerpts quoted below are taken from CommonMark Spec Version 0.30, which is by now over four years old.

From section 4.5 Fenced Code Blocks:

A code fence is a sequence of at least three consecutive backtick characters (`) or tildes (~). (Tildes and backticks cannot be mixed.)

The content of the code block consists of all subsequent lines, until a closing code fence of the same type as the code block began with (backticks or tildes), and with at least as many backticks or tildes as the opening code fence.

From section 6.1 Code Spans:

A backtick string is a string of one or more backtick characters (`) that is neither preceded nor followed by a backtick.

A code span begins with a backtick string and ends with a backtick string of equal length. The contents of the code span are the characters between these two backtick strings, normalized in the following ways:

First, line endings are converted to spaces.

If the resulting string both begins and ends with a space character, but does not consist entirely of space characters, a single space character is removed from the front and back. This allows you to include code that begins or ends with backtick characters, which must be separated by whitespace from the opening or closing backtick strings.

I hope these little nuggets of Markdown trivialities will one day prove useful in your own Markdown misfortunes.

Read on website | #web | #technology

Susam's Web Posts

I Will Not Add Query Strings to Your URLs

Contents

Wisdom on the Web

Wander on the Web

Misfeature

Broken URLs

Qualms

Conclusion

Wander Console 0.6.0

From RSS to Atom

Contents

Impulse Coding

Atom Entries

Temporary Workaround

Does It Matter?

References

QuickQWERTY 1.2.3

Apr '26 Notes

Contents

Coset Results

Subgroup

Cosets

Coset Membership

Coset Equality

Partitioning a Group into Disjoint Cosets

Computing

Wander Console Updates

Flush Output

Other Stuff

QuickQWERTY 1.2.2

Wander Console 0.5.0

Wander Console 0.4.0

Contents

Wildcard Patterns

The 'via' Query Parameter

Console Picker Algorithm

Allow Links that Open in New Tab

Community

Wander Console 0.3.0

Wander Console 0.2.0

Wander Console 0.1.0

Wander Console

HN Skins 0.4.0

HN Skins 0.3.0

HN Skins 0.2.0

HN Skins 0.1.0

Nerd Quiz #4

QuickQWERTY 1.2.1

Attention Media ≠ Social Networks

Nested Code Fences in Markdown

Contents

Basic Code Fences

Fancy Code Fences

Basic Code Spans

Fancy Code Spans

Specification