For the desperate people whose tests all just broke, I have a page on my public endpoint testing server that exactly reproduces the classic design: https://example.testserver.host/. You can just use update the URL and go back to blissful ignorance.
Before this submission, it genuinely never occurred to me that people would actually have tests that rely on example.com being up, let alone depend on its content.
Opus 5.5 in Claude code added a unit test checking DNS and internet access using example.com as the target on a project I'm building yesterday. It's an agent sandbox (I know, yet another), so the call was expected to fail, but still... It should have at least targeted a captive portal endpoint or a project owned domain.
I caught it in review but thousands of others won't.
More likely the agents will fix it as fast as they made it and people will continue not caring so much about reviewing what agents are doing for cases where they don't care so much (which is a lot of cases)
I have been naively using example.com in the browser. WHATWG URL throws if there is no origin when constructing the URL. Web apps that need to construct _just_ the pathname must go through this song and dance. `new URL("/blah", "https://example.com").pathname` This isn't relying on any external example.com service but I'm learning the lesson.
So is wrong to use captive.apple.com for tests as well? It's nice to know your network can reach the internet and these pages are generally fairly reliable.
A HTTP request for http://www.msftncsi.com/ncsi.txt returning 200 OK and the text Microsoft NCSI. This is the url windows has been using for decades to determine if the device is online. It is highly reliable.
These pages are unfortunately getting special treatment from quite a few networks these days, among other things to avoid iOS users getting stuck in a loop of trying to connect, and getting disconnected by the OS due to a "non-working connection" for local-only networks like in-flight entertainment systems.
It's a shame that we don't have any standards for the concerns of canonically testing Internet reachability and for authoritatively redirecting network users to a captive payment portal (without having to resort to ugly hacks that often break with TLS).
If they say that it's not intended as a reliable service for testing purposes it can't be that reliable, and they might take it down or change the structure arbitrarily at any time, etc.
So yeah I would pick something simpler for a network access test probably?
The only risk is that Apple hasn't announced it as being reliable or promised to keep it up or the format the same in any way. They could 100% pull it out from under everyone, with modifications in their OSes prepared for the change
What kind of tests would this break? Aside from the advice on the page itself (which, iirc, is new anyway), i'm wondering why any testing would actually involve the contents of design of the page. Just a weird 'sanity' check?
Our saas had an internal-use, undocumented, publicly accessibly but not publicly used (by us) health endpoint that included an internal version number.
We removed the version and a few customers complained we broke their stuff.
I landed a massive database upgrade on a payroll system years ago, end-to-end standardization, cleanup, and improvement with a seamless rollout.
Our back clapping was interrupted by an angry customer phone call. One of our customers had gotten access to our internal schema and had a massive reporting setup in Access solving most of his needs.
I ended up converting his reports using some internal tools we had. Customers will grab anything to give them.
[The caliber and depth of his reports were such I wish we would have bought them instead of making me spend months building out our own reports.]
I had one customer that managed to figure out which stored procs were tied to which reports we were using. Then he would deliberately break it in weird subtle ways them and call tech support 'just to see how we would react'.
While not the same thing, I was once stung by a test failing because a dependency of a dependency decided on a minor version bump that foo@example.com wasn't a valid email address.
One of the critical things I use example.com for is triggering wifi login portals as it doesn't not use SSL. Certificate pinning and all that fancy stuff has made it so the browser has an absolutely awful time figuring out what to do when https://google.com does not present itself as the same host it was earlier.
Same server also has an endpoint for that: http://no-tls.testserver.host. Sends RST for any attempted TLS connections, so clients always fall back to plain HTTP.
Also as a _long_ list of other specialist TLS endpoint configurations if you're interested, which can be arbitrarily combined, see https://testserver.host/#tls-endpoints.
That gives neat tricks like https://tls-v1-2--expired--incomplete-chain--http2.testserve...: only accepts TLS 1.2, then sends an expired certificate but fails to send the intermediate cert for the chain (so the client must infer it) and then negotiates HTTP/2 for the connection on top. Fun!
> For the desperate people whose tests all just broke...
The web page at example.com is maintained as a courtesy by IANA in order to explain the purpose of the example.com domain to wayward humans.
In the nomenclature of RFC 2119, one MUST NOT design computer systems that rely on the correct operation of an HTTP server at that domain. [0] Plus, it's _really_ rude to pound on a small-scale service being provided as a courtesy... go hit the home page of a tech megacorp (such as Microsoft or Google) or the status page for a major CDN (such as Cloudflare or Akamai) instead!
[0] I expect that someone here will want to pop up with a "gotcha" where they say something like "Oh, but IANA's email says that automated use is strongly recommended against, rather than prohibited and besides, they can't actually stop me from doing it!". To that, I reply "Sure, and standards-writers can't actually stop implementers that do the profoundly antisocial thing and do the things they MUST NOT. As any adult who's been paying attention to the world around them throughout their lives knows, there's only so much you can do to stop people who are very determined to be enormous assholes.".
That they've put it behind Cloudflare doesn't mean it's "served by Cloudflare"
Maybe Cloudflare gives them a good rate, or is donating service, but it's also possible or even likely that IANA is just using Cloudflare as a vendor, and therefore is paying for all the traffic, which is why they don't want people relying on it or hammering it all the time
In the nomenclature of RFC 2119, one MUST NOT design computer systems that rely on the correct operation of an HTTP server at [example.com.] Plus, it's *_really_* rude to pound on a small-scale service being provided as a courtesy.
> go hit the home page of a tech megacorp (such as Microsoft or Google)
Much too big for metered data, full of ads, and importantly they all mandate HTTPS these days, which breaks my use case of forcing a captive portal on paid/login-gated Wi-Fi to render.
example.com is (unfortunately for the IATA) the almost perfect "can I reach the Internet" service: It's unlikely to go away, supports HTTP, is fairly small, easy to remember, and I don't care if a captive portal poisons my DNS cache with a fake response temporarily.
I think we're on the verge of a big disruption coming for the example.com's business. I can see a SaaS opportunity. We can use AI to speed up time to market.
example.com-as-a-service, although XaaS is no longer hip so now it'll be example.ai, an AI that will generate an example.com style landing page using frontier AI models. $20 / month for the beginner plan (100 requests/month), contact us for pricing.
Maybe, but in any case far fewer than will rely on a fixed page. I think very clearly it'd be a net-pain lowering move for example.com to randomize the results a bit, if they felt like it.
~~I feel like that would probably meaningfully increase the amount of compute necessary to serve the website. As it were it’s likely static html, but any randomness would have to rely on a server side operation for each request and couldn’t be cached.~~
I went and checked and in fact it is already serving a script s.js so I guess you’re right!
… then you'll be back in the world before CloudFlare set up its public DNS service, which is also the world where taking over 1.1.1.1 in the first place came at the price of having to measure how many people had ignorantly or lazily put four ones into somewhere that they had to fill in an IP address, and how much bogus and positively risky (to them) traffic there was being caused by people just thinking 'I will just use 1.1.1.1, which is not a real IP address.'.
Just before COVID, a random unidentifiable person reported that 1.1.1.1 saw 60Mb/s of ICMP echo traffic. When APNIC first experimented with making it a valid network almost a decade before that, it was being sent 50Mb/s of general IP traffic. Amusingly, a side-effect of what CloudFlare has done is that it has stopped all of that traffic leaving the edges of the Internet, and funnelling in to one big central place.
There are those who remember the risks of 'just use 1.1.1.1' and how the people who did that used be characterized as 'bad Internet citizens'. The ServerFault answer (q.v.) is from 2011.
That was part of Cloudflare's pitch to take over the 1.1.1.1 address block - that Cloudflare had the relatively unique ability to handle all the junk traffic as anycast nearby to any source, not overloading the wider internet.
I don’t use this one, because I remember when long, long ago it was in some random IP address block belonging to someone, and this particular address didn’t reply to ping.
Likely for good reason, even back then there would have bound to be lots of misconfigured endpoints trying to access 1.1.1.1, so just blocking it at the earliest point possible kept at lot of the annoyance away. Nowadays, it’s an anycast address, so it’s slightly less bad.
I've already seen a couple of train hotspots respond locally to pings directed at 1.1.1.1 and 8.8.8.8, apparently in effort to convince devices that they are on a good network event if the internet itself is broken due to being in the tunnel.
Pinging such an address is inherently a troublesome practice. This address, like many public DNS servers (resolvers as well as root and authoritative ones), uses "anycast" routing methodology.
Pinging an anycast address will yield a cornucopia of different results. Of course, people who naïvely "ping" a recognizable or easy-to-type IPv4 address get what they deserve, especially when they enshrine it into software, unit tests, or the LLM coughs up such tokens on their behalf.
Fundamentally, the question is "what do you really want to test?" by pinging a particular IPv4? Do you want to test Layer 3 connectivity? Test your ISP's backbone and connectivity? Test only your upstream router? Test the existence of ICMP in your stack and theirs?
... the possibilities are endless. Your router can do anything. Ping zombo.com.
Why is pinging an anycast address an issue? Most people, myself included, ping addresses like 1.1.1.1 to check for internet connectivity. It's a perfectly valid check, you don't need a fine grained test all the time. If it fails, then I go looking in detail. Most of the time it will pass.
Pinging a domain name could fail for a variety of other reasons, particularly if it's not a reliable site. Cloudflare's business is being reliable; few sites would be a better choice.
The problem is pinging 1.1.1.1 can end up hitting a really really close-by server. You could be having ISP issues and 1.1.1.1 could end up being closer to you than the issue is, so you get okay ping results but still unable to access resources on the other side of the issue
You're thinking far too hard. If I can ping 1.1.1.1, it means traffic is flowing from inside my network, to outside my network. I'm not going to diagnose my ISP's issues, that's their job, I just need to know that it's not my problem.
You can have routing issues that alow some IP addresses to work and others not, anycast or not. It's not supposed to be a comprehensive connectivity report, just a sanity check that the intertubes are connected at all.
Of course. Ping 1.1.1.1 is not a test that everything is working great. DNS could be dead. IPv6 could be black holed.
But it’s a good “is traffic making it past my router with some semblance of connectivity” sanity check, and easier than finding the provider-side next hop address.
The corollary is that when it fails, you've no idea what failed, or how it failed, or if the "fail" is even valid or relevant, because you're abusing a service that isn't designed for you and isn't fit for use. So if it succeeds, you really don't know, because perhaps your train's WiFi router is responding with a spoofed IPv4 address while it goes through the Chunnel. And if it fails, you really don't know, because you DGAF about learning formal troubleshooting methods and couldn't be arsed to find out your upstream router's address in order to simply isolate your PING to a singular link, rather than relying on complex routing rules, especially those introduced by "anycast". And perhaps you can manually dig in when your manual checks fail, but your AI agent or automated script DGAF about your nuanced knowledge that 1.1.1.1 isn't your upstream router; in fact it's not yours at all and has nothing to do with your network topology. But at least it looks elegant.
I ping 1.1.1.1 to see if my internet is working or not after a couple of websites don’t load. I reboot the router. I keep the terminal where I’m pinging 1.1.1.1 open until I start seeing responses and then I know my internet is back. Then I continue my web browsing and other online activities.
Fun fact, ping is the standard windows way to wait a given number of seconds[0]. You're supposed to ping 127.0.0.1, but I saw a lot of scripts pinging 1.1.1.1 or 8.8.8.8
Don't see why this is relevant, ping is not traceroute or http. If you just want to test if packets can leave your network then its a good enough test and anycast doesn't affect that.
that's a really good point I'd not considered but if you just hash the response and the followups remain consistent (they do in this case)... you're gonna do just fine.
Yes, the revised version that alternated through the translations was posted a week ago on Monday. It was revised to no longer do that on Friday based on reasonable feedback that it wasn't the best idea for accessibility. Now all the translations are presented together, but your preferred language will be bumped to the top based on your browser language preference.
Oh that's good, I had that complaint when it first changed. It was difficult to read the entire chunk of text before it transitioned to a different language. And then who knows how long it took to loop back round to english
Aww, I was wondering if I was just looking at an older version. How disappointing. But I suppose, unsurprising given the conversation here. I'm sure such transitions would be bringing a surprising amount of instrumentation down.
It still probably is as it is now. But there's better arguments to support localization than transitions.
> For some reason the localization still requires javascript
Well, no. The purpose of the redesign is to move as much of the page content as possible into an external javascript file. There's no question of what parts of the page require javascript; the question is what parts of the page can be moved into javascript.
Yeah apologies, bad wording on my part. I really just meant the removal of the opening line which made it clear that it was a correspondence: "I am happy to try and answer any questions you have."
The hilariously ironic thing is that Dunk, a Google Engineer, went out of their way to make a hyperlink of "example dot com" so that people could tap and visit the site, even where Davies' email specifically reiterates that the domain is for documentation purposes only and nobody should ever be actually visiting the site!
McCarthy of DebugBear clearly noticed this admonition and perhaps the mistake as well, because they have carefully highlighted "example.com" without linking to it, although they apparently have visited the site in order to copy and cite screenshots and unpack the CSS, and describe an extensive history of it.
Kinda interesting that this is getting so much attention.
Ultimately, the example.com domain is just there so you can use it in documentation or code examples. Your expectation about the actual example.com domain should basically be none at all, maybe with the exception that it's hopefully not doing actively malicious things like serving malware or running a catchall mailbox feeding spammers.
Whatever HTML they server or if they serve anything at all... shouldn't matter.
This has rendered my usecase for example.org worthless.
I used this website several times per week to clean my glasses as it provided a nearly completely white website - optimal for allowing me to spot smudges on my glasses. It is now black by default for me, or grey.
You also failed to make it a link. In order to automatically create links, you should use the format https://example.com/ with scheme included. So in your case you'd do https://example.com/
Thanks. But you also forgot to include the link. You used the example domain https://example.com/ but forgot to substitute the correct address of the website. In this case that would be https://example.com/
By typing it or copy+pasting it, which is much more time consuming and much less convenient.
(I actually tried the select, share [sic], open in browser route, but for some reason I only saw opera as an option, not chrome, the browser I actually use for everything. Links still rule.)
That’s because they are in code blocks. Oh I hate people who abuse code blocks for things that aren’t code. Domain names are not code. Your browser doesn’t display URLs in a monospace coding font. Stop this abuse.
> There is no requirement there is a HTTP service present on the host in order to fulfill its purpose, we just operate it as a courtesy.
I guess this is also done to prevent bad actors from abusing the fact that this domain is hit by people who might not know what they're doing (the ones copy-pasting code without reading it)
More the DNS and DNSSEC and the like. Whether or not there is actually an HTTP server responding is irrelevant to whether or not those securely point anywhere but a malicious system.
It's just part and parcel of owning the domain. It's one thing to have the domain, but the next step is offering an active website so that people that actually put that domain in a browser can see it's a placeholder.
When you own a domain, you can choose not to serve anything on HTTP/S, and still nobody can come and serve some other website on your domain, because they don’t control the DNS records.
HTTP is used by billions of people while SSH is not. It should be pretty apparent. Why do people answer my phone calls and texts but everyone is ignoring my smoke signals?
I appreciate that there must be a lot of traffic hitting this web server so maybe every little bit of savings counts, but I kinda wish the HTML wasn't so heavily compressed.
I remember getting my start with web development by simply reading the source code of sites I found interesting and trying to recreate them. example.com isn't exactly very interesting but it feels like compressing the HTML isn't in the spirit of things either.
On a side note I'd love to see how complicated (or maybe simple) the hosting setup for this site actually is behind the scenes.
It's static content in Cloudflare CDN. Seems unlikely that many requests actually hit any kind of origin server. If I were building something like this the origin would be Cloudflare R2 (or similar) object storage, with a very long cache lifetime. Nothing to update, nothing to maintain.
But beyond that, what are you referring to as compression? When I look at the source I see a VERY short easily human-readable HTML page. It doesn't have newlines, but I wouldn't consider that compression. It also references a short, but non-obfuscated javascript file at https://example.com/s.js .
I'm not sure I believe that IANA's goal of reducing serving costs is met by a site where each letter is wrapped in a <span>...
separate shower thought: Microsoft have used contoso.com as the example domain in their docs for 25+ years. I bet the logs for that domain are absolutely wild.
So they made a prior tweak to reduce fetches of favicon.ico ( by link rel="icon" href="data:," )
but why do they still serve the overlarge meaningless html page to requests for robots.txt ?!
Given that they have gotten spiked by automated requests lately, aka agents, wouldn't this at least be respected by the big players?
Also, as they are using Cloudflare, wouldn't a discussion (or a note from IANA or CF) about how they configured CF request rejection and how one should love CF's flat rate static serving be appropriate?
It’s not incompetence they are intentionally doing as much as possible via javascript to reduce their bandwidth costs. This is because most access to the domain is automated (not using a browser, using curl etc) and doesn’t load javascript
Really funny coincidence, I noticed amazon wasn't loading a couple hours ago. Thought it was my internet so went to example.com to check and noticed the redesign. Figured it was a while ago though.
I wonder if it's injecting both through javascript to minimize breaking people's tests. Trying to reduce dealing with non-ascii characters and svg tags.
Reducing egress bandwidth, if you have automated clients that don't support JS you save precious bytes from being served by not embedding the SVG, which at example.com scale may be worth it
Why does this site need JS at all? It appears to inject all of the other languages via client side JS. Why can't they all just be served from the server?
You may want to check your settings again or clear you cache as there is definitely a `s.js`[0] file that inserts the `<svg>` element into the HTML DOM[1], which you can see does not contain the element itself.
0
```
var B = document.body, P, p;
B.children[0].insertAdjacentHTML("afterend", "<p lang=ar dir=rtl>هذا النطاق مُخصص للاستخدام في أمثلة التوثيق دون الحاجة إلى إذن. هذه ليست خدمة، يُرجى تجنب الاعتماد عليها لأغراض الاختبار والمراقبة.</p><p lang=zh>该域名仅用于文档示例，无需获得许可。这并非一项服务，请勿将其用于测试和监控目的。</p><p lang=fr>L’usage de ce domaine est réservé à des exemples de documentation, sans autorisation préalable. Il ne s’agit pas d’un service ; son utilisation à des fins de test ou de surveillance est à éviter.</p><p lang=ru>Данный домен предназначен для использования в примерах документации без необходимости получения предварительного разрешения. Это не сервис; не рекомендуется его использование для тестирования и мониторинга.</p><p lang=es>Este dominio está destinado al uso en ejemplos de documentación sin necesidad de permiso. Esto no es un servicio; evitar utilizarlo para realizar pruebas o monitoreos.</p><a href=https://iana.org/help/example-domains>Learn more</a>");
P = [...B.querySelectorAll("p")];
P[0].lang = "en";
navigator.languages.some(l => p = P.find(p => p.lang == l.split("-")[0]));
p = p || P[0];
B.prepend(p);
B.insertAdjacentHTML("afterbegin", '<style>svg{display:block;margin:-2.75em auto 0;opacity:.55}p+p{font-size:.875em;opacity:.6}</style><svg viewBox=0,0,20,20 width=44 height=44 fill=currentColor aria-hidden=true><path fill=none stroke=currentColor stroke-width=1.3 stroke-linejoin=round d="M6 4H3v12q4 0 7 1.5-1-3.5-4-4.5V2q3.5 1 4 3v12.5q3-1.5 7-1.5V4q-4.5 0-7 1"/><g transform=rotate(-6,13.6,8.3)><path id=q d="M11.5 6.6h1.8v1.8l-1 1.6-.6-.3.8-1.3h-1z"/><use href=#q x=2.4 /></g></svg>')
body {
font: 16px/1.6 system-ui,sans-serif;
max-width: 26em;
margin: auto;
padding: 25vh 2em 2em;
text-align: center
}
</style>
</head>
<body>
<p>This domain is for use in documentation examples without needing permission. This is not a service; avoid relying on it for testing and monitoring purposes.</p>
<script src=/s.js></script>
</body>
It seems that they changed it again and now there is minified HTML with english message + simple script that injects the other languages and the icon, no animation. Which to me seems much more sensible.
I did not know about that empty favicon trick, and I will be adding it to all my base HTML templates for my Flask projects where I get annoyed at all the 404s scrolling past as my browser keeps requesting a favicon that will not exist in that form.
I actually noticed that! I was debugging some HTTP proxy and typed out an HTTP/1.1 request by hand in netcat, and was wondering why I saw a giant blob of dynamic nonsense instead of a small bit of markup.
(This was when it had the language scroll though, I think moving the extras to JS should have resolved that issue; maybe I was misremembering the details too.)
example.com is not a user-facing service. I don't want to disparage the designer, but it's not meant to be pretty. It's meant to be small so you can copy its response to an automated test or visually verify its correctness, or just out of the bandwidth consideration you'd still be serving it to billions of requestors, even if you insist it's not meant to be used that way.
I never queried it _automatically_, but many projects do. They might not have set out to provide that use, but what happened has happened, and breaking this implicit contract for a silly redesign is an odd choice.
> and was wondering why I saw a giant blob of dynamic nonsense instead of a small bit of markup.
It's 4.5x the size of the original. Which sounds bad, until you notice this means it's 2540 bytes and serves the explanatory text in 6x the number of languages. And it's now served with brotli compression, for a total of 1713 bytes. Or 334 bytes if you only request the HTML (which still has a usable page with the full English version of the explanatory text), like a basic scraper or a health check would. So for the vast majority of traffic, it's now half the size of the old version.
I notice it's been through several revisions, the pre-2025 version of the site most people are familiar with is 1270 bytes because it didn't make use of any minification, and it also triggered an unnecessary `/favicon.ico` request because it didn't use the trick to cancel the request.
These are what these companies' respective browsers/operating systems use to check if they have an internet connection and are not stuck behind a captive portal. Given the enormous quantity of devices out in the world that are relying on these URLs to stay up, you can probably feel comfortable using these in your tests. If they do go down, good chance you'll see news articles about it before you notice your tests failing.
Cloudflare is the latest in a number of different CDN providers we've used to serve this over many years. I guess the question is, why is it important not to use a CDN?
Same reason everyone else uses CF. Static page CDN only became too cheap to meter ~16 years ago, but if CF was around in the 90s IANA would have used it in the 90s too
Same here! Yes, I used it as the well-known site that supported unencrypted http and no HSTS. Yes, I was one of those guys using the site "as a service" rather than simply documentation. Admittedly, it was perhaps interchangeable with other sites, but since it reliably worked for that purpose, I couldn't be arsed to find other ones.
Isn't there a dedicated site called nohttps or something? NeverSSL? I just tried http NeverSSL, and it redirected me to an https page with a random hostname and an Amazon SSL certificate!
Does example.com still function this way, with http and no HSTS? I noticed that my Chrome browser was immediately redirected in the customary way.
No, NeverSSL changed to have SSL a while back - around the time that Chrome/Firefox started throwing big warning signs and/or blocking non-https pages.
Wait, are you saying neverssl.com now sometimes uses SSL?
If this is a joke, it's over my head. If not, I'm not knowledgeable enough in this space to pass judgement, but that would seem crazy to me. Like the owner should consider a domain name change :)
It has SSL on the main domain, and some subdomains without SSL (wildcard perhaps?). The main domain (when accessed over SSL?) redirects to a subdomain with plain HTTP.
Kinda weird setup!
Edit: okay, the subdomains also have SSL. I guess the random subdomain thing is to make sure it hasn’t been cached in a “this site has HTTPS” list. The HTTPS is needed, of course, to please the browser makers, as the parent says.
Wow finally, a "no ssl" site that actually doesn't use SSL! I was so disappointed when "NEVERssl" added SSL, like you literally have ONE job. Some clients automatically upgrade http->https or prepend https if left unspecified, so it's nice to have an explicitly HTTP-only (IE port 80-only) service for testing/captive portals.
example.com, example.net, example.org, and example.edu, along with .example, are all classified as the same type of name, reserved for documentation purposes.
Furthermore, there are IPv4 and IPv6 addresses that are reserved for documentation purposes! You never need to write live addresses, or even private RFC 1918 addresses!
Nonsense. One may very well explain to a newbie how to link to an external site using example.com as target. Though I would not do it, with that layout. I would link to Google probably.
We (IPinfo) have a practice of adopting low-maintenance, high-utility services. Generally, we adopt websites that would have become defunct and run them as services.
Even though IANA says example.com is intended for documentation purposes and not as a service, I feel that it nonetheless serves as a service to the broader internet. I think there are some cornerstone services on the internet, regardless of the maintainers' intentions or their legal definitions.
For example, we ourselves sometimes have to recognize that we are not a standard API service. Considering that we expect to process 3 trillion requests this year, a major outage could take down a good chunk of IT systems everywhere. So, we invested in infrastructure to avoid outages. We then started adopting other cornerstone services and running them indefinitely because we might as well support the users who depend on them because we have infrastructure to support them and us.
Not an advert. I wish we ran example.com. However, we run a few other services.
The issue with example.com is this statement:
> This domain is for use in documentation examples without needing permission. This is not a service; avoid relying on it for testing and monitoring purposes.
The problem is that, disclaimer or not, if a website provides a useful utility, people will use it as a service.
If example.com breaks, IANA can reasonably say it was never intended to be stable. But that does not change the fact that people will use it for testing and monitoring.
What we do is adopt legacy services whose maintainers can no longer afford, want, or have the resources to maintain them, and invest in keeping them running reliably.
Example.com is one of the sites I visit most often for troubleshooting my connection on the go. It's great for tickling a captive Wi-Fi login that didn't trigger properly.
I think what you describe is exactly right, it has organically become a service and it is maintained with that in mind. That doesn't mean you shouldn't be clear about what its for and what should be avoided. Otherwise it is setting up an implicit contract that it will service those needs without limitation which turns it into an even bigger dependency.
Not to mention indefinitely. Any decommission, even decades in the future, will be messy. Or imagine if ipinfo got bought up by private equity and they turned off the free service, or maybe attached some sort of agent-pay token to the response[0] so that they collect revenue from running it when possible.
This feels like a “falsehood developers think about localization.”
It’s always funny when I get French YouTube ads. As if YouTube is desperately trying to offload the adbuys anywhere they can by pretending that Canada must mean French.
I hate this with a passion. Google is very guilty of this. Their AI summary very often insists on replying in the (IP) local language, nevermind the language of my HTTP headers, Google profile, or the actual natural language request...
I want to read content only in a specific language. This is different language than the language of the country I live in. I have my operating system, my keyboard layout, my browser, my accept languages header, everything configured to the language I want.
And get I often get websites and ads in the language of the country my IP points to
I think the whole point is for it to be a static site that is easily distributed with little-to-no overhead, as it probably receives a non-trivial amount of traffic.
Interesting, didn’t know that. Also I would have expected German and Hindi to be official UN languages given the number of native speakers in the world
For the desperate people whose tests all just broke, I have a page on my public endpoint testing server that exactly reproduces the classic design: https://example.testserver.host/. You can just use update the URL and go back to blissful ignorance.
Also open-source and self-hostable: https://github.com/httptoolkit/testserver. Lots of other endpoints too, full docs here: https://testserver.host/
If you do have tests relying on it... stop it:
> This is not a service; avoid relying on it for testing and monitoring purposes.
Before this submission, it genuinely never occurred to me that people would actually have tests that rely on example.com being up, let alone depend on its content.
Opus 5.5 in Claude code added a unit test checking DNS and internet access using example.com as the target on a project I'm building yesterday. It's an agent sandbox (I know, yet another), so the call was expected to fail, but still... It should have at least targeted a captive portal endpoint or a project owned domain.
I caught it in review but thousands of others won't.
That ship has sailed I would say.
If it breaks in those thousands of projects, that will be an important lesson to review your agents more carefully. So a net win?
More likely the agents will fix it as fast as they made it and people will continue not caring so much about reviewing what agents are doing for cases where they don't care so much (which is a lot of cases)
I have been naively using example.com in the browser. WHATWG URL throws if there is no origin when constructing the URL. Web apps that need to construct _just_ the pathname must go through this song and dance. `new URL("/blah", "https://example.com").pathname` This isn't relying on any external example.com service but I'm learning the lesson.
So is wrong to use captive.apple.com for tests as well? It's nice to know your network can reach the internet and these pages are generally fairly reliable.
[1]: https://captive.apple.com/
A HTTP request for http://www.msftncsi.com/ncsi.txt returning 200 OK and the text Microsoft NCSI. This is the url windows has been using for decades to determine if the device is online. It is highly reliable.
These pages are unfortunately getting special treatment from quite a few networks these days, among other things to avoid iOS users getting stuck in a loop of trying to connect, and getting disconnected by the OS due to a "non-working connection" for local-only networks like in-flight entertainment systems.
It's a shame that we don't have any standards for the concerns of canonically testing Internet reachability and for authoritatively redirecting network users to a captive payment portal (without having to resort to ugly hacks that often break with TLS).
RFC8910, from 2020: https://www.rfc-editor.org/rfc/rfc8910.html
Woah, amazing! Do you know if any popular network stacks actually support it?
But even just having an RFC to yell about is great, thank you :)
If they say that it's not intended as a reliable service for testing purposes it can't be that reliable, and they might take it down or change the structure arbitrarily at any time, etc.
So yeah I would pick something simpler for a network access test probably?
Just don't blame the user and/or their internet service when the service you're relying on inevitably goes down.
Why not ping 8.8.8.8 or 1.1.1.1 or time.windows.com? Things that are actually designed to be highly available services
Is there any reason to believe captive.apple.com is any less "designed to be highly available" than time.windows.com?
It's a static page that every Apple device relies on saying only:
The only risk is that Apple hasn't announced it as being reliable or promised to keep it up or the format the same in any way. They could 100% pull it out from under everyone, with modifications in their OSes prepared for the change
1.1 is enough.
Thanks! I'd forgotten that 1.1 and 1.1.1 resolve. I'll not infrequently ping 1.1.1.1.1 with overzealous typing. That will, alas, not resolve.
What kind of tests would this break? Aside from the advice on the page itself (which, iirc, is new anyway), i'm wondering why any testing would actually involve the contents of design of the page. Just a weird 'sanity' check?
People depend on all sorts of weird stuff working exactly as expected. https://www.hyrumslaw.com/
As an example, depending on "man -w" not outputting anything to stderr: https://unix.stackexchange.com/questions/405783/why-does-man...
Our saas had an internal-use, undocumented, publicly accessibly but not publicly used (by us) health endpoint that included an internal version number.
We removed the version and a few customers complained we broke their stuff.
I landed a massive database upgrade on a payroll system years ago, end-to-end standardization, cleanup, and improvement with a seamless rollout.
Our back clapping was interrupted by an angry customer phone call. One of our customers had gotten access to our internal schema and had a massive reporting setup in Access solving most of his needs.
I ended up converting his reports using some internal tools we had. Customers will grab anything to give them.
[The caliber and depth of his reports were such I wish we would have bought them instead of making me spend months building out our own reports.]
> [The caliber and depth of his reports were such I wish we would have bought them instead of making me spend months building out our own reports.]
Learning from customers is some of the best learning you can do. :)
I had one customer that managed to figure out which stored procs were tied to which reports we were using. Then he would deliberately break it in weird subtle ways them and call tech support 'just to see how we would react'.
The API surface is what’s exposed, not just what’s documented.
This is probably an even more important principle in the age of LLMs.
While not the same thing, I was once stung by a test failing because a dependency of a dependency decided on a minor version bump that foo@example.com wasn't a valid email address.
Probably akin to: https://xkcd.com/1172/
I knew what this was before clicking on the link, but for a lot of people, it may be a 1053.
One of the critical things I use example.com for is triggering wifi login portals as it doesn't not use SSL. Certificate pinning and all that fancy stuff has made it so the browser has an absolutely awful time figuring out what to do when https://google.com does not present itself as the same host it was earlier.
I like using http://neverssl.com/ for that... easy to remember (although I guess example.com was as well).
Same server also has an endpoint for that: http://no-tls.testserver.host. Sends RST for any attempted TLS connections, so clients always fall back to plain HTTP.
Also as a _long_ list of other specialist TLS endpoint configurations if you're interested, which can be arbitrarily combined, see https://testserver.host/#tls-endpoints.
That gives neat tricks like https://tls-v1-2--expired--incomplete-chain--http2.testserve...: only accepts TLS 1.2, then sends an expired certificate but fails to send the intermediate cert for the chain (so the client must infer it) and then negotiates HTTP/2 for the connection on top. Fun!
Do you do any sort of rate-limiting to stop people/bots from hammering it (accidentally or otherwise)? Or do you just let it fall over under load.
> For the desperate people whose tests all just broke...
The web page at example.com is maintained as a courtesy by IANA in order to explain the purpose of the example.com domain to wayward humans.
In the nomenclature of RFC 2119, one MUST NOT design computer systems that rely on the correct operation of an HTTP server at that domain. [0] Plus, it's _really_ rude to pound on a small-scale service being provided as a courtesy... go hit the home page of a tech megacorp (such as Microsoft or Google) or the status page for a major CDN (such as Cloudflare or Akamai) instead!
[0] I expect that someone here will want to pop up with a "gotcha" where they say something like "Oh, but IANA's email says that automated use is strongly recommended against, rather than prohibited and besides, they can't actually stop me from doing it!". To that, I reply "Sure, and standards-writers can't actually stop implementers that do the profoundly antisocial thing and do the things they MUST NOT. As any adult who's been paying attention to the world around them throughout their lives knows, there's only so much you can do to stop people who are very determined to be enormous assholes.".
> the status page for a major CDN (such as Cloudflare or Akamai)
Would you settle for any old static page served by Cloudflare? For instance, example.com? https://bgp.tools/dns/example.com
That they've put it behind Cloudflare doesn't mean it's "served by Cloudflare"
Maybe Cloudflare gives them a good rate, or is donating service, but it's also possible or even likely that IANA is just using Cloudflare as a vendor, and therefore is paying for all the traffic, which is why they don't want people relying on it or hammering it all the time
> go hit the home page of a tech megacorp (such as Microsoft or Google)
Much too big for metered data, full of ads, and importantly they all mandate HTTPS these days, which breaks my use case of forcing a captive portal on paid/login-gated Wi-Fi to render.
example.com is (unfortunately for the IATA) the almost perfect "can I reach the Internet" service: It's unlikely to go away, supports HTTP, is fairly small, easy to remember, and I don't care if a captive portal poisons my DNS cache with a fake response temporarily.
I think we're on the verge of a big disruption coming for the example.com's business. I can see a SaaS opportunity. We can use AI to speed up time to market.
/s
example.com-as-a-service, although XaaS is no longer hip so now it'll be example.ai, an AI that will generate an example.com style landing page using frontier AI models. $20 / month for the beginner plan (100 requests/month), contact us for pricing.
Enterprise ready. Coming soon in Q4.
I'm wondering how many automated tests this change broke.
Yes, I know it's not the example.com's fault, and it's a side effect of Hyrum's Law:
> This is not a service, avoid relying on it for testing and monitoring purposes.
Don't we all have a few fragile tests?
That's an excellent reason to change the page design occasionally, so that people learn not to rely on it (the hard way if necessary).
Go a step further and randomize it. Different page at every request.
But then people will rely on the randomness to pick their quicksort pivots or something like that.
Maybe, but in any case far fewer than will rely on a fixed page. I think very clearly it'd be a net-pain lowering move for example.com to randomize the results a bit, if they felt like it.
~~I feel like that would probably meaningfully increase the amount of compute necessary to serve the website. As it were it’s likely static html, but any randomness would have to rely on a server side operation for each request and couldn’t be cached.~~
I went and checked and in fact it is already serving a script s.js so I guess you’re right!
If 1.1.1.1 ever stops responding to pings…
… then you'll be back in the world before CloudFlare set up its public DNS service, which is also the world where taking over 1.1.1.1 in the first place came at the price of having to measure how many people had ignorantly or lazily put four ones into somewhere that they had to fill in an IP address, and how much bogus and positively risky (to them) traffic there was being caused by people just thinking 'I will just use 1.1.1.1, which is not a real IP address.'.
Just before COVID, a random unidentifiable person reported that 1.1.1.1 saw 60Mb/s of ICMP echo traffic. When APNIC first experimented with making it a valid network almost a decade before that, it was being sent 50Mb/s of general IP traffic. Amusingly, a side-effect of what CloudFlare has done is that it has stopped all of that traffic leaving the edges of the Internet, and funnelling in to one big central place.
There are those who remember the risks of 'just use 1.1.1.1' and how the people who did that used be characterized as 'bad Internet citizens'. The ServerFault answer (q.v.) is from 2011.
* https://labs.ripe.net/author/franz/pollution-in-18/
* https://news.ycombinator.com/item?id=19335833
* https://serverfault.com/a/339782
That was part of Cloudflare's pitch to take over the 1.1.1.1 address block - that Cloudflare had the relatively unique ability to handle all the junk traffic as anycast nearby to any source, not overloading the wider internet.
I don’t use this one, because I remember when long, long ago it was in some random IP address block belonging to someone, and this particular address didn’t reply to ping.
Likely for good reason, even back then there would have bound to be lots of misconfigured endpoints trying to access 1.1.1.1, so just blocking it at the earliest point possible kept at lot of the annoyance away. Nowadays, it’s an anycast address, so it’s slightly less bad.
But for me, old habits die hard.
I've already seen a couple of train hotspots respond locally to pings directed at 1.1.1.1 and 8.8.8.8, apparently in effort to convince devices that they are on a good network event if the internet itself is broken due to being in the tunnel.
If 1.1.1.1 fails, I will try 8.8.8.8
> 1.1.1.1
Pinging such an address is inherently a troublesome practice. This address, like many public DNS servers (resolvers as well as root and authoritative ones), uses "anycast" routing methodology.
https://en.wikipedia.org/wiki/Anycast
Pinging an anycast address will yield a cornucopia of different results. Of course, people who naïvely "ping" a recognizable or easy-to-type IPv4 address get what they deserve, especially when they enshrine it into software, unit tests, or the LLM coughs up such tokens on their behalf.
Fundamentally, the question is "what do you really want to test?" by pinging a particular IPv4? Do you want to test Layer 3 connectivity? Test your ISP's backbone and connectivity? Test only your upstream router? Test the existence of ICMP in your stack and theirs?
... the possibilities are endless. Your router can do anything. Ping zombo.com.
Why is pinging an anycast address an issue? Most people, myself included, ping addresses like 1.1.1.1 to check for internet connectivity. It's a perfectly valid check, you don't need a fine grained test all the time. If it fails, then I go looking in detail. Most of the time it will pass.
Pinging a domain name could fail for a variety of other reasons, particularly if it's not a reliable site. Cloudflare's business is being reliable; few sites would be a better choice.
The problem is pinging 1.1.1.1 can end up hitting a really really close-by server. You could be having ISP issues and 1.1.1.1 could end up being closer to you than the issue is, so you get okay ping results but still unable to access resources on the other side of the issue
You're thinking far too hard. If I can ping 1.1.1.1, it means traffic is flowing from inside my network, to outside my network. I'm not going to diagnose my ISP's issues, that's their job, I just need to know that it's not my problem.
Most of my connection issues are between the street and my computer, I don't think Cloudflare got there. Yet.
You can have routing issues that alow some IP addresses to work and others not, anycast or not. It's not supposed to be a comprehensive connectivity report, just a sanity check that the intertubes are connected at all.
Of course. Ping 1.1.1.1 is not a test that everything is working great. DNS could be dead. IPv6 could be black holed.
But it’s a good “is traffic making it past my router with some semblance of connectivity” sanity check, and easier than finding the provider-side next hop address.
The corollary is that when it fails, you've no idea what failed, or how it failed, or if the "fail" is even valid or relevant, because you're abusing a service that isn't designed for you and isn't fit for use. So if it succeeds, you really don't know, because perhaps your train's WiFi router is responding with a spoofed IPv4 address while it goes through the Chunnel. And if it fails, you really don't know, because you DGAF about learning formal troubleshooting methods and couldn't be arsed to find out your upstream router's address in order to simply isolate your PING to a singular link, rather than relying on complex routing rules, especially those introduced by "anycast". And perhaps you can manually dig in when your manual checks fail, but your AI agent or automated script DGAF about your nuanced knowledge that 1.1.1.1 isn't your upstream router; in fact it's not yours at all and has nothing to do with your network topology. But at least it looks elegant.
I ping 1.1.1.1 to see if my internet is working or not after a couple of websites don’t load. I reboot the router. I keep the terminal where I’m pinging 1.1.1.1 open until I start seeing responses and then I know my internet is back. Then I continue my web browsing and other online activities.
I’m so lazy I just type `ping 1.1` and it still works (apparently 1.0.0.1 is also always up).
It's much faster just pinging 127.0.0.1 for such purposes. Try it!
In the spirit of M4v3R's post, that should be 127.1 . (-:
ping 0 would save you 3 bytes!
Same
Any IP address could one day change where it is being announced from, or become anycast, or change the number of hops between you and it.
I don't think people are using `ping 1.1.1.1` as a stable API, rather as a yes/no test of the network segment that they control.
Fun fact, ping is the standard windows way to wait a given number of seconds[0]. You're supposed to ping 127.0.0.1, but I saw a lot of scripts pinging 1.1.1.1 or 8.8.8.8
[0]: https://stackoverflow.com/questions/1672338/how-to-sleep-for...
> standard windows way
The thread you linked to suggests 'timeout'[0]
[0] https://learn.microsoft.com/en-us/windows-server/administrat...
Scroll down. Ping is preferred because timeout is buggy
>Of course, people who naïvely "ping" a recognizable or easy-to-type IPv4 address get what they deserve
Look what happened because of what you did, what it led to! Two microservices are in critical condition and you're laughing. You're laughing.
Why would it matter?
https://utcc.utoronto.ca/~cks/space/blog/tech/InternetPathsM...
Don't see why this is relevant, ping is not traceroute or http. If you just want to test if packets can leave your network then its a good enough test and anycast doesn't affect that.
that's a really good point I'd not considered but if you just hash the response and the followups remain consistent (they do in this case)... you're gonna do just fine.
> there's a gradual opacity transition
Looks like they removed this and instead just show all languages with no CSS animation now.
Yes, the revised version that alternated through the translations was posted a week ago on Monday. It was revised to no longer do that on Friday based on reasonable feedback that it wasn't the best idea for accessibility. Now all the translations are presented together, but your preferred language will be bumped to the top based on your browser language preference.
Oh that's good, I had that complaint when it first changed. It was difficult to read the entire chunk of text before it transitioned to a different language. And then who knows how long it took to loop back round to english
Aww, I was wondering if I was just looking at an older version. How disappointing. But I suppose, unsurprising given the conversation here. I'm sure such transitions would be bringing a surprising amount of instrumentation down.
It still probably is as it is now. But there's better arguments to support localization than transitions.
For some reason the localization still requires javascript... (I'm pretty sure the transition could have been done in just CSS as well)
> For some reason the localization still requires javascript
Well, no. The purpose of the redesign is to move as much of the page content as possible into an external javascript file. There's no question of what parts of the page require javascript; the question is what parts of the page can be moved into javascript.
Discussed here a few days ago: https://news.ycombinator.com/item?id=49915060
Thanks! Macroexpanded:
IANA's email about why example.com changed - https://news.ycombinator.com/item?id=49915060 - Sept 2026 (20 comments)
And the "statement" in this article is the text from that email, disingenuously edited. I'm not sure why, but it rubs me the wrong way.
What makes you think it's edited?
Yeah apologies, bad wording on my part. I really just meant the removal of the opening line which made it clear that it was a correspondence: "I am happy to try and answer any questions you have."
The hilariously ironic thing is that Dunk, a Google Engineer, went out of their way to make a hyperlink of "example dot com" so that people could tap and visit the site, even where Davies' email specifically reiterates that the domain is for documentation purposes only and nobody should ever be actually visiting the site!
McCarthy of DebugBear clearly noticed this admonition and perhaps the mistake as well, because they have carefully highlighted "example.com" without linking to it, although they apparently have visited the site in order to copy and cite screenshots and unpack the CSS, and describe an extensive history of it.
Kinda interesting that this is getting so much attention.
Ultimately, the example.com domain is just there so you can use it in documentation or code examples. Your expectation about the actual example.com domain should basically be none at all, maybe with the exception that it's hopefully not doing actively malicious things like serving malware or running a catchall mailbox feeding spammers.
Whatever HTML they server or if they serve anything at all... shouldn't matter.
This has rendered my usecase for example.org worthless. I used this website several times per week to clean my glasses as it provided a nearly completely white website - optimal for allowing me to spot smudges on my glasses. It is now black by default for me, or grey.
I am devastated.
Do the kids not know about about:blank?
Wouldn't help OC if their system is dark mode.
It's not as easy to remember, but you can also do:
The elites don't want you to know this, but you can disable dark mode in your settings. I actually use light mode every day, but keep that between us.
whitedisplay.com can be your new friend. Just select the preferred color.
> Just select the preferred color.
I kinda expected this to be Ford Model T "Any color the customer wants, as long as it’s black" joke.
Good old Hyrum's Law
Relevant xkcd: https://xkcd.com/1172/
The page mentions example.com 14 times and not a single one is a link
You also failed to make it a link. In order to automatically create links, you should use the format https://example.com/ with scheme included. So in your case you'd do https://example.com/
Guess I am part of the problem. Thank you for the link kind stranger.
Thanks. But you also forgot to include the link. You used the example domain https://example.com/ but forgot to substitute the correct address of the website. In this case that would be https://example.com/
Outrageous. How will people ever find it?!
By typing it or copy+pasting it, which is much more time consuming and much less convenient.
(I actually tried the select, share [sic], open in browser route, but for some reason I only saw opera as an option, not chrome, the browser I actually use for everything. Links still rule.)
For me, it's double click and cmd-U which launches the Search with Google service on macOS.
That’s because they are in code blocks. Oh I hate people who abuse code blocks for things that aren’t code. Domain names are not code. Your browser doesn’t display URLs in a monospace coding font. Stop this abuse.
> There is no requirement there is a HTTP service present on the host in order to fulfill its purpose, we just operate it as a courtesy.
I guess this is also done to prevent bad actors from abusing the fact that this domain is hit by people who might not know what they're doing (the ones copy-pasting code without reading it)
More the DNS and DNSSEC and the like. Whether or not there is actually an HTTP server responding is irrelevant to whether or not those securely point anywhere but a malicious system.
Yeah there’s definitely a lot of sensitive data that gets sent to the domain just because of people not changing configs
I don't understand, what risk would there be if they chose _not_ to serve a site?
Someone else could
No, because they can't register the domain, it's already taken, no matter if a web service is running behind it or not
It's just part and parcel of owning the domain. It's one thing to have the domain, but the next step is offering an active website so that people that actually put that domain in a browser can see it's a placeholder.
This is not the point discussed here.
> Someone else could
When you own a domain, you can choose not to serve anything on HTTP/S, and still nobody can come and serve some other website on your domain, because they don’t control the DNS records.
Why HTTP? Why not SSH? When I try to connect to it that way, I get no response. How will I know it's a placeholder?
HTTP is used by billions of people while SSH is not. It should be pretty apparent. Why do people answer my phone calls and texts but everyone is ignoring my smoke signals?
What is the smoke signal for “we’ve been trying to reach you about your car’s extended warranty“?
I appreciate that there must be a lot of traffic hitting this web server so maybe every little bit of savings counts, but I kinda wish the HTML wasn't so heavily compressed.
I remember getting my start with web development by simply reading the source code of sites I found interesting and trying to recreate them. example.com isn't exactly very interesting but it feels like compressing the HTML isn't in the spirit of things either.
On a side note I'd love to see how complicated (or maybe simple) the hosting setup for this site actually is behind the scenes.
It's static content in Cloudflare CDN. Seems unlikely that many requests actually hit any kind of origin server. If I were building something like this the origin would be Cloudflare R2 (or similar) object storage, with a very long cache lifetime. Nothing to update, nothing to maintain.
But beyond that, what are you referring to as compression? When I look at the source I see a VERY short easily human-readable HTML page. It doesn't have newlines, but I wouldn't consider that compression. It also references a short, but non-obfuscated javascript file at https://example.com/s.js .
Maybe I misunderstood what you're referring to...
I'm not sure I believe that IANA's goal of reducing serving costs is met by a site where each letter is wrapped in a <span>...
separate shower thought: Microsoft have used contoso.com as the example domain in their docs for 25+ years. I bet the logs for that domain are absolutely wild.
> Since most of the traffic to the site is automated it tends not to fetch the Javascript file, reducing overall data requirements.
Well, IATA – give me api.example.com or sniff my user agent and I'll be out of your hair :)
Here is some context: https://kimdavies.com/being-exemplary
This is mine now! <link rel=icon href=data:,>
I put this on all my websites that don’t have a favicon. I hate the default browser behavior of making a request to /favicon.ico without being asked.
I’m curious… who hosts and controls this content? IANA? Never really considered this question until now…
see https://news.ycombinator.com/item?id=42705284 for some historic detail on this
the 'about' link on example.com sends you here https://www.iana.org/help/example-domains
which does appear to say that yes, IANA hosts this
So they made a prior tweak to reduce fetches of favicon.ico ( by link rel="icon" href="data:," ) but why do they still serve the overlarge meaningless html page to requests for robots.txt ?!
Given that they have gotten spiked by automated requests lately, aka agents, wouldn't this at least be respected by the big players?
Also, as they are using Cloudflare, wouldn't a discussion (or a note from IANA or CF) about how they configured CF request rejection and how one should love CF's flat rate static serving be appropriate?
Am I missing something?
It's quite sad to see incompetence at these levels. This is a static page that loads JS. They even inject styles via JS...
It’s not incompetence they are intentionally doing as much as possible via javascript to reduce their bandwidth costs. This is because most access to the domain is automated (not using a browser, using curl etc) and doesn’t load javascript
They use Cloudflare, which does not charge for bandwidth.
For free users. But cloudflare picks and chooses its free users.
It's been about 10 years since I last saw this site and accidentally visited it a few days ago and thought "I don't remember it looking like this"
Really funny coincidence, I noticed amazon wasn't loading a couple hours ago. Thought it was my internet so went to example.com to check and noticed the redesign. Figured it was a while ago though.
Maybe this is the most web-development thing possible: even example.com eventually became a frontend project.
> Along with introducing multi-language support, the JavaScript code also inserts an SVG book icon.
Wait - using JS for dynamic content is understandable, but why using it for inserting a static SVG?
I wonder if it's injecting both through javascript to minimize breaking people's tests. Trying to reduce dealing with non-ascii characters and svg tags.
Reducing egress bandwidth, if you have automated clients that don't support JS you save precious bytes from being served by not embedding the SVG, which at example.com scale may be worth it
Wouldn't removing the pointless SVG image entirely save the most precious bytes from being served no matter if JS (or SVG) is supported or not?
It's good to have a little color/flavor/spice in life.
Ah yes because bog standard corporate design is how you add color/flavor/spice in life.
Judging by the reaction to one SVG, it's the most they could do! Can you imagine if it was some art nouveau project instead?
Good for them, and it's still better than nothing. Without it the page would almost looked broken it's so bare.
Why does this site need JS at all? It appears to inject all of the other languages via client side JS. Why can't they all just be served from the server?
Simple answer is the article is wrong, I tried it with disabled JS and still see the SVG.
The whole article reads like something put together with AI, so maybe it’s not too surprising.
I get nothing but a gray page with the following with JS disabled https://i.imgur.com/81aYPeB.png in Firefox via javascript.enabled set to false.
You may want to check your settings again or clear you cache as there is definitely a `s.js`[0] file that inserts the `<svg>` element into the HTML DOM[1], which you can see does not contain the element itself.
0
```
var B = document.body, P, p; B.children[0].insertAdjacentHTML("afterend", "<p lang=ar dir=rtl>هذا النطاق مُخصص للاستخدام في أمثلة التوثيق دون الحاجة إلى إذن. هذه ليست خدمة، يُرجى تجنب الاعتماد عليها لأغراض الاختبار والمراقبة.</p><p lang=zh>该域名仅用于文档示例，无需获得许可。这并非一项服务，请勿将其用于测试和监控目的。</p><p lang=fr>L’usage de ce domaine est réservé à des exemples de documentation, sans autorisation préalable. Il ne s’agit pas d’un service ; son utilisation à des fins de test ou de surveillance est à éviter.</p><p lang=ru>Данный домен предназначен для использования в примерах документации без необходимости получения предварительного разрешения. Это не сервис; не рекомендуется его использование для тестирования и мониторинга.</p><p lang=es>Este dominio está destinado al uso en ejemplos de documentación sin necesidad de permiso. Esto no es un servicio; evitar utilizarlo para realizar pruebas o monitoreos.</p><a href=https://iana.org/help/example-domains>Learn more</a>"); P = [...B.querySelectorAll("p")]; P[0].lang = "en"; navigator.languages.some(l => p = P.find(p => p.lang == l.split("-")[0])); p = p || P[0]; B.prepend(p); B.insertAdjacentHTML("afterbegin", '<style>svg{display:block;margin:-2.75em auto 0;opacity:.55}p+p{font-size:.875em;opacity:.6}</style><svg viewBox=0,0,20,20 width=44 height=44 fill=currentColor aria-hidden=true><path fill=none stroke=currentColor stroke-width=1.3 stroke-linejoin=round d="M6 4H3v12q4 0 7 1.5-1-3.5-4-4.5V2q3.5 1 4 3v12.5q3-1.5 7-1.5V4q-4.5 0-7 1"/><g transform=rotate(-6,13.6,8.3)><path id=q d="M11.5 6.6h1.8v1.8l-1 1.6-.6-.3.8-1.3h-1z"/><use href=#q x=2.4 /></g></svg>')
```
1
```
<!doctype html> <html lang=en> <head> <meta charset=utf-8> <link rel=icon href=data:,> <meta name=viewport content="width=device-width,initial-scale=1"> <title>Example Domain</title> <style> html { color-scheme: light dark; background: light-dark(#eee,#222) }
</html>```
``` doesn’t work on HN, prepend two spaces to each line to get a code block.
Regardless, it’s a lot of code, so maybe better move it out to a pastebin/gist?
It seems that they changed it again and now there is minified HTML with english message + simple script that injects the other languages and the icon, no animation. Which to me seems much more sensible.
I did not know about that empty favicon trick, and I will be adding it to all my base HTML templates for my Flask projects where I get annoyed at all the 404s scrolling past as my browser keeps requesting a favicon that will not exist in that form.
I actually noticed that! I was debugging some HTTP proxy and typed out an HTTP/1.1 request by hand in netcat, and was wondering why I saw a giant blob of dynamic nonsense instead of a small bit of markup.
(This was when it had the language scroll though, I think moving the extras to JS should have resolved that issue; maybe I was misremembering the details too.)
example.com is not a user-facing service. I don't want to disparage the designer, but it's not meant to be pretty. It's meant to be small so you can copy its response to an automated test or visually verify its correctness, or just out of the bandwidth consideration you'd still be serving it to billions of requestors, even if you insist it's not meant to be used that way.
I believe your use is what they're trying to discourage.
example.com is meant to be legally allowed to be used in text such as "You visit a website (example.com) to browse the internet"
It is not meant to be queried by automated tests or used as a service.
I never queried it _automatically_, but many projects do. They might not have set out to provide that use, but what happened has happened, and breaking this implicit contract for a silly redesign is an odd choice.
> and was wondering why I saw a giant blob of dynamic nonsense instead of a small bit of markup.
It's 4.5x the size of the original. Which sounds bad, until you notice this means it's 2540 bytes and serves the explanatory text in 6x the number of languages. And it's now served with brotli compression, for a total of 1713 bytes. Or 334 bytes if you only request the HTML (which still has a usable page with the full English version of the explanatory text), like a basic scraper or a health check would. So for the vast majority of traffic, it's now half the size of the old version.
I notice it's been through several revisions, the pre-2025 version of the site most people are familiar with is 1270 bytes because it didn't make use of any minification, and it also triggered an unnecessary `/favicon.ico` request because it didn't use the trick to cancel the request.
> It's meant to be small [...], even if [...] it's not meant to be used that way
It is not, in fact, _meant_ to be used that way.
What should I use for example URLs in my testa.
Google operates http://connectivitycheck.gstatic.com/generate_204 and http://www.gstatic.com/generate_204, both of which return a 204 No Content response.
Apple operates http://captive.apple.com, which returns `<HTML><HEAD><TITLE>Success</TITLE></HEAD><BODY>Success</BODY></HTML>`.
Microsoft operates http://www.msftconnecttest.com/connecttest.txt, which returns `Microsoft Connect Test`.
Mozilla operates http://detectportal.firefox.com, which returns `success`.
These are what these companies' respective browsers/operating systems use to check if they have an internet connection and are not stuck behind a captive portal. Given the enormous quantity of devices out in the world that are relying on these URLs to stay up, you can probably feel comfortable using these in your tests. If they do go down, good chance you'll see news articles about it before you notice your tests failing.
I just wonder how much unintended traffic they receive.
eg If they put DNS, NTP etc on it... Likely billions per hour or something equally ludicrous.
Deluge is likely an exponential understatement.
who's job was it to make incremental stylesheet changes on example.com in the 2010s?
The government. It was a top secret skunkworks project to see if centering a <div> was possible.
I see the government continues to try solving impossible problems in the shadows. A pity.
Whose. Who's is a contraction of who + is
I just don't understand why the operator, IANA, has to use Cloudflare for example.com
It cannot manage a one page website. WTF
There are no page "aseets", no need for images, CSS, etc. It was a text-only website to test connectivity
iana.org forwards to Cloudflare, too
Fortunately the FTP server service still works
Without assistance from a third party
If it's been a sensible decision to use Cloudflare for some period of time then why did IANA wait until now
(Yes, I know they used Akamai in the past)
Cloudflare is the latest in a number of different CDN providers we've used to serve this over many years. I guess the question is, why is it important not to use a CDN?
HN is currently anti Cloudflare. The only thing worse probably would’ve been to serve the static files on GitHub.io
Same reason everyone else uses CF. Static page CDN only became too cheap to meter ~16 years ago, but if CF was around in the 90s IANA would have used it in the 90s too
IANA has a history of hosting even more important services with third-parties [0] [1], so this doesn't seem too problematic to me.
[0]: https://root-servers.org/
[1]: https://datatracker.ietf.org/doc/html/rfc6305.html
NB. IANA isn't responsible for "root servers"
It's only responsible for root.zone, root.hints, .arpa and .int zones
AS112 is a volunteer project not a company
If it's a name in com/net/org only meant for documentation then why serve a page there for decades. If traffic is a problem then take it offline
Years ago IANA started requiring a user-agent header
FTP access to root.zone remains
*assets
Well, it's a popular website. By using Cloudflare I think they will save a lot of bandwidth and traffic because the page and assets can be cached.
Ive been using example1.com recently for availability type testing for what its worth.
it's nice, and it surprised me when I saw the redesign.... I use it to navigate into captive wifi portals that always seem to be misconfigured.
I've used http://www.com for that.
Same here! Yes, I used it as the well-known site that supported unencrypted http and no HSTS. Yes, I was one of those guys using the site "as a service" rather than simply documentation. Admittedly, it was perhaps interchangeable with other sites, but since it reliably worked for that purpose, I couldn't be arsed to find other ones.
Isn't there a dedicated site called nohttps or something? NeverSSL? I just tried http NeverSSL, and it redirected me to an https page with a random hostname and an Amazon SSL certificate!
Does example.com still function this way, with http and no HSTS? I noticed that my Chrome browser was immediately redirected in the customary way.
http://neverssl.com/online/ seems to work for me?
Are you perhaps on an untrusted network which is spoofing that url to make it redirect somewhere else?
No, NeverSSL changed to have SSL a while back - around the time that Chrome/Firefox started throwing big warning signs and/or blocking non-https pages.
Wait, are you saying neverssl.com now sometimes uses SSL?
If this is a joke, it's over my head. If not, I'm not knowledgeable enough in this space to pass judgement, but that would seem crazy to me. Like the owner should consider a domain name change :)
It has SSL on the main domain, and some subdomains without SSL (wildcard perhaps?). The main domain (when accessed over SSL?) redirects to a subdomain with plain HTTP.
Kinda weird setup!
Edit: okay, the subdomains also have SSL. I guess the random subdomain thing is to make sure it hasn’t been cached in a “this site has HTTPS” list. The HTTPS is needed, of course, to please the browser makers, as the parent says.
http://http.rip/ is one I use
Wow finally, a "no ssl" site that actually doesn't use SSL! I was so disappointed when "NEVERssl" added SSL, like you literally have ONE job. Some clients automatically upgrade http->https or prepend https if left unspecified, so it's nice to have an explicitly HTTP-only (IE port 80-only) service for testing/captive portals.
This is very useful, thanks for sharing!
My browser (iOS safari) just takes me to the ssl version from that link, so “never ssl” seems wrong?
http://example.com/ works for me in Safari, warning that the site is “not secure”, but only if I type it in by hand.
Andddd it broke all my EVALs... ;)
the copy on example.com looks llm-generated now...
And example.net!
https://en.wikipedia.org/wiki/Example.com
example.com, example.net, example.org, and example.edu, along with .example, are all classified as the same type of name, reserved for documentation purposes.
Furthermore, there are IPv4 and IPv6 addresses that are reserved for documentation purposes! You never need to write live addresses, or even private RFC 1918 addresses!
https://en.wikipedia.org/wiki/IPv4#Special-use_addresses
Previously:
IANA's email about why example.com changed
https://news.ycombinator.com/item?id=49915060
I find the text shown on this site passive agressive.
When a newbie learning programming sees this message, then this is a downer.
"Hello World from example.com [more info]"
I find it pretty neutral, if maybe a bit bureaucratic. It explains what this domain does, and what it’s not supposed to be used for – that’s it.
BTW, a "background:white;" for the body improves things. But still, the versions from 2013 or 2019 are much better in my opinion.
I think the previous versions didn’t really have a goal to discourage people from relying on example.com as infrastructure? *shrug*
>When a newbie learning programming
A newbie shouldn't test their programs using example.com
Nonsense. One may very well explain to a newbie how to link to an external site using example.com as target. Though I would not do it, with that layout. I would link to Google probably.
Static test is still, IMHO best, but if they want that fancy transition they could at least have used a gif. All that js is terrible overkill.
The language translation animation only lasted a few days. Now it’s static with English at the top and translations below.
It's English at the top for you. The top language depends on your browser settings afaik.
You understand that the GIF would be larger than the JS, right? (Apart from other problems with this idea.)
Overkill for a web of humans loading sites in browsers. Not overkill for a dead internet with swarms of agents DDoSing everything in sight.
It's down?
We (IPinfo) have a practice of adopting low-maintenance, high-utility services. Generally, we adopt websites that would have become defunct and run them as services.
Even though IANA says example.com is intended for documentation purposes and not as a service, I feel that it nonetheless serves as a service to the broader internet. I think there are some cornerstone services on the internet, regardless of the maintainers' intentions or their legal definitions.
For example, we ourselves sometimes have to recognize that we are not a standard API service. Considering that we expect to process 3 trillion requests this year, a major outage could take down a good chunk of IT systems everywhere. So, we invested in infrastructure to avoid outages. We then started adopting other cornerstone services and running them indefinitely because we might as well support the users who depend on them because we have infrastructure to support them and us.
For a moment I thought you were saying ipinfo runs example.com.
Instead it just seems to be an advert for ipinfo on a post about example.com?
Not an advert. I wish we ran example.com. However, we run a few other services.
The issue with example.com is this statement:
> This domain is for use in documentation examples without needing permission. This is not a service; avoid relying on it for testing and monitoring purposes.
The problem is that, disclaimer or not, if a website provides a useful utility, people will use it as a service.
If example.com breaks, IANA can reasonably say it was never intended to be stable. But that does not change the fact that people will use it for testing and monitoring.
What we do is adopt legacy services whose maintainers can no longer afford, want, or have the resources to maintain them, and invest in keeping them running reliably.
My question was rhetorical ;)
Example.com is one of the sites I visit most often for troubleshooting my connection on the go. It's great for tickling a captive Wi-Fi login that didn't trigger properly.
neverssl.com is slightly better for that, since it resists caching.
I think what you describe is exactly right, it has organically become a service and it is maintained with that in mind. That doesn't mean you shouldn't be clear about what its for and what should be avoided. Otherwise it is setting up an implicit contract that it will service those needs without limitation which turns it into an even bigger dependency.
Not to mention indefinitely. Any decommission, even decades in the future, will be messy. Or imagine if ipinfo got bought up by private equity and they turned off the free service, or maybe attached some sort of agent-pay token to the response[0] so that they collect revenue from running it when possible.
0: https://blog.cloudflare.com/monetization-gateway-beta/
IANA, a natural monopoly, can't even host simple html page is kinda funny.
IANA and ICANN should be moved under UN ITU from a non profit, so that they don't make breaking changes without approval.
What would have been nice is that depending on the visitors IP address’s region, showing a localised message instead of fixed number of languages
This feels like a “falsehood developers think about localization.”
It’s always funny when I get French YouTube ads. As if YouTube is desperately trying to offload the adbuys anywhere they can by pretending that Canada must mean French.
I hate this with a passion. Google is very guilty of this. Their AI summary very often insists on replying in the (IP) local language, nevermind the language of my HTTP headers, Google profile, or the actual natural language request...
Do you have ca-fr (or whatever) in your accept language header?
I want to read content only in a specific language. This is different language than the language of the country I live in. I have my operating system, my keyboard layout, my browser, my accept languages header, everything configured to the language I want.
And get I often get websites and ads in the language of the country my IP points to
Google famously think they know better than you what languages you want.
I think the whole point is for it to be a static site that is easily distributed with little-to-no overhead, as it probably receives a non-trivial amount of traffic.
Yep, that's exactly their reasoning:
> As it tends to be heavily trafficked, the overall bandwidth is a key consideration
Then they should have just kept it english only without any script.
That assumes that (i) every region has a primary language, and (ii) everyone in that region speaks that language, both of which are false.
(And false often enough that guessing language based on IP address works badly in practice, I hear.)
Just check the Accept-Language header
I believe it's just using the official languages of the UN
Interesting, didn’t know that. Also I would have expected German and Hindi to be official UN languages given the number of native speakers in the world
Guess most of German speaking and Hindi speaking also know English.