[geofft]

They can, but they believe in rough consensus and working code and in being conservative in what they do and liberal in what they accept from others.

It's clear that a lot of real-world HTTP implementations are using HTTP 418 to mean I'm A Teapot, whether or not they should, and that reassigning it would cause practical difficulties.

[peterwwillis]

That's not clear to me. Nobody relies on the feature, it's a tiny piece of code, only some vendors support it, and it's not even part of the spec. (It also can't be reassigned because it was never assigned to begin with)

If they like 418, they can add it to the spec. But this idea that it's been "consumed" is weird. It's like saying <blink> was "consumed" in HTML because two browsers used to support it. It was just an unofficial feature of popular vendor software, and is now officially disavowed.

Basically what we're saying is we're not going to implement 418 in the spec, but at the same time, we're not going to let anyone use it?

[JoshTriplett]

> It's like saying <blink> was "consumed" in HTML because two browsers used to support it.

Which is accurate. Reintroducing a tag named "blink" into HTML to mean something other than "blinking text" would be a really bad idea for compatibility reasons.

[peterwwillis]

So we're trying to avoid breaking the internet by committing to people expecting an error when they're talking to a teapot? So that past implementations of clients connecting to teapots won't break?

Like I say, we could also just add it to the spec... But that would be ridiculous, because why would you support telling someone you were a teapot in an HTTP protocol, when the feature is only useful in a protocol for brewing coffee? That would be like taking a programming language's syntax for assigning data objects and turning it into it's own standard for a text file format.

Maybe a time traveler set this whole thing up 19 years ago just to troll people on standards boards.

[phab]

> That would be like taking a programming language's syntax for assigning data objects and turning it into it's own standard for a text file format.

Looking at you, JSON...

[vacri]

It's a miniscule bit of humour that consumes 1 of 31 used slots in a 100-slot block, in an area of tech that isn't even remotely drowning in 'fun stuff'. What does it really matter? There are plenty of perfectly serious RFCs that are even less implemented than 418.

It's not like this RFC ratifying a 20-year-old joke that 'made it' is going to generate a cavalcade of lookalikes.

[pwdisswordfish]

You never know. Some people used to think 4 billion slots would be more than enough for addressing every host in the world, and so they allocated the addresses haphazardly. Now look at the sorry state of IPv6 adoption 20 years after it's been designed.

Some people used to think 65536 code points would be more than enough to encode every character in every language. Later it turned out not to be the case. Expanding that space necessitated the creation of hacks of various degrees of awfulness like UTF-16, CESU-8, WTF-8; today nearly every Unicode-aware environment has to be prepared to deal with unpaired surrogates somehow.

Expanding a fixed-size namespace is a pain. Who knows if HTTP status codes won't become scarce some day. They better not be wasted on frivolities.

[dhosek]

Having been around for the creation of Unicode, there were plenty of people who argued that 16 bits was insufficient for encoding all the characters and fought for the competing ISO standard which was 32 bits. A final compromise made Unicode a subset of ISO-10646 and now the two character sets are synchronized and act as parallel standards. UTF-8/16/32 were part of things from the beginning not because of the 16-bit assumption behind Unicode 1.0, but for the sake of keeping data reasonably sized. UTF-8 allows for 7-bit ascii files (which were the majority of text files back in the day) to be compliant without change and kept the default size of text files from being doubled or quadrupled in an era in which that was a serious concern for both bandwidth (dial-up connections at 2400baud or lower were quite common) and storage (20MB was considered a HUGE amount of storage).

[quuquuquu]

As frivolous as you think 418 is, people use it.

You honestly can't, in this situation, be the judge of what is "frivolous use" and what is not.

[dexterdog]

Except blink is a string that means what it is. 418 is one of a small set of codes to use and has no purpose as it exists other than nonsense.

[vidarh]

Several discussions have included anecdotes from people about how they are using the 418 code for various things, so I don't know how you can be sure that nobody relies on it.

[concede_pluto]

Say you want to add some feature critical to correctness. "Request can't be committed until resent to other quorum members." Something serious like that. You can't use 418 for that feature because some non-conforming servers are already returning it as a joke. You'd have to triage new RFCs until you get one that isn't important enough to work reliably, and sacrifice it to use up the tainted code.

[vacri]

The 400-block is 69% free, after 28 years of development. If we're ever in the position where we need to free up just one more code, then we're already consuming beyond our capacity and will need to sort out an alternative anyway.

[vidarh]

Exactly. If we run out of 400-codes without an alternative having been set out, such as e.g. designating one or more of the remaining 400 codes to specify extended formats for, somebody is not doing their jobs.

E.g. once (if ever) we're down to 10 or so free ones, assign them as broad categories and specify that the response text needs to contain an additional, longer status code (so, say, "499 1234 some text" and "499 4242 some text").

It's not as if tech doesn't have a long history of including extension fields like that, and it still works with the existing grammar