Hacker News Reader

It's hilarious that browsers will use loosy goosy parsing on the HTML of a web page but strictly interpret a JPEG to which its img tag refers.

Why? Why do car rentals typically not require cash up front, but hotel rentals universally do? The economics are similar. Sometimes it's simple path dependency. Something is a certain way because it's always been that way and it'd be too expensive to change it now.

At least browsers all use the same loosy goosy HTML parsing now. It was hell on earth when each browser had its own error recovery strategy. In a sense, there is no longer any such thing as an invalid HTML document: every code point sequence has a canonical interpretation now.

It’s Postel’s law at the end of the day: “be conservative in what you do, be liberal in what you accept from others”. As a site owner I want my site to fail loudly and quickly before a user sees a problem; as a user I never want to see a problem.

ePub is in a nice place: the number of documents to check for errors is reasonable, and the resulting artefect is designed to be shipped and never (or rarely) amended. That means that we can shift the balance towards strict parsing. But for a web site of thousands (or millions) of documents that are being amended regularly, the balance shifts back to loose parsing as the best way of meeting user needs.

Postel's Law sounds nice but it can result in major problems. It results in a de facto spec that differs from the written spec, and disagreements about what a piece of data actually means can lead to bugs and even security vulnerabilities.

Having strictly parsed HTML from the start would be fine. You'd check it before you ship it and you'd make sure it's valid.

Requiring it now would be a disaster, of course. There's so much malformed HTML out there. But making HTML parsers accept garbage at the beginning was the wrong choice.

The widespread acceptance of Postel’s Law also encourages poor authorship, because if you know clients have to be liberal in what they accept, there is no incentive to be conservative in what you send.

Isn't the developer always the first user? With strict parsing, testing a site before launch would show you the problem right there and allow you to fix it to launch a bug free site.

What about a 5 year old client hitting a new server or the reverse? Is the only solution just don't do that?

That is what the HTML 4 !doctype Tag - https://www.quackit.com/html/html_4/tags/doctype_html_public... - was meant for. Ideally, an old client that cannot parse a newer HTML spec should inform the user to update the client. Developers could then either support old versions of their site (more easily, in such a case) or can simply decide not to deal with outdated browsers. (In fact, isn't that what most BiGTech sites actually do now, through convoluted browser feature detections - "Your browser is no longer supported ...").

Postel's law has turned out to be good for prototypes and experiments, but a horrible way to make anything with any degree of reliability.

What about evolving standards in a system that must handle clients or servers which implement anything from tomorrows feature today to 10 years prior. Shouldn't failures be as graceful as possible?

No, now or 10 years later, syntax error (like unclosed tag or missing quote) is a syntax error. Ignoring it is always a bad idea.

Unknown tags / attributes, on the other hand, may be ignored for future compatibility.. but I believe XHTML ignores them already.

My favorite is how this interacts with the oh so fun mistake many people make of adding a `<div/>` thinking they are doing it right.

Especially since it is correct in JSX, adding to the confusion.

It doesn't help that self closing/void tags are actually a thing in HTML. input, image, meta, br-- these can be misleading to a naive developer. You can try to say something about how a div requires text content, but empty divs get used structurally by design teams all the time in practice...

Yeah, but these are self-closing. Writing "<br/>" is actually wrong and just corrected to "<br>" in HTML. HTML does not have that syntax, that would be XML. And that is why "<div/>" is corrected to "<div>" and the closing "</div>" is inserted somewhere.

The problem is that you are collapsing two users with very different needs into a single one.

1. If you are authoring an XHTML file, yes, you want the renderer to be as picky as possible and yell loud and clear if you make a mistake. This helps you produce a valid, well-formed document.

2. If you are an end user reading an XHTML file, it's not your file, it's not your fault if there are bugs, and there's jack shit you can do to fix it. You just want the browser to do its best to show you something reasonable so you can read the page and get on with your life.

XHTML optimizes for 1 at the expense of 2. HTML5 optimizes for 2 at the expense of 1.

For some reason generating a valid wire format seems to be no problem for people when it comes to json. Forgot to escape a quote? Woops, that’s on me, should have used a serializer.

But add a few angled braces in there and lord have-a mercy, ain’t nobody can understand this ampersand mumbo jumbo, I wanna hand write my documents and generate wutever, yous better jus deal with it gosh dangit.

I prefer the current situation too but I still think it’s funny somehow we just never bought into serializers for html. Maybe the idea was before its time? I’m sure you’d have no such parsing problems in the wild if you introduced JTML now. Clearly people know how to serialize.

> For some reason generating a valid wire format seems to be no problem for people when it comes to json.

The "some reason" is that JSON is most often produced by code, written by programmers.

That's not the case for HTML which is often hand-authored directly by non-programmer users. Most people on Earth literally don't know what the word "syntax" means, much less feel comfortable dealing with syntax errors in text files.

"Most people on Earth" can't write HTML. And if one doesn't know what the word "syntax" means, they are not going to have much success writing HTML - after all, ignoring the errors does not make them go away.

Instead, it seems your choices are either:

(1) errors cause random changes on page, such as text squishing into small box, or whole paragraphs missing, or random words appearing on page, or style being applied to wrong text etc...

(2) errors cause page to fail loading with no text shown at all

Both of those are pretty bad, but I wish web would go with (2) instead of (1). Because in that case, we'd have whole ecosystem of tooling appear... Imagine "auto fixup" feature in web servers which "fixes up" bad HTML into good one. For basic users, it looks the same as today. But instead of fixup and guesswork being done by users' browsers (which author has no control of) it would be done by author's webhosts (which author can upgrade or not upgrade as needed...)

You are vastly underestimating how much of the web was and is successfully authored by non-technical people who barely understand the difference between a word processor and a text editor but are able to cobble together enough HTML tags that it mostly looks like they want and lets them share their knowledge with the rest of the world.

> 2. If you are an end user reading an XHTML file, it's not your file, it's not your fault if there are bugs, and there's jack shit you can do to fix it. You just want the browser to do its best to show you something reasonable so you can read the page and get on with your life.

Here's the thing though, if all XHTML clients are strict about it then that means the content is broken for EVERYONE which presumably means it gets noticed pretty quickly as long as the site is being maintained by anyone.

Compare that to HTML where if a page is doing something wrong but in a way that happens to work in Webkit/Blink while it barfs all over the place in Gecko it could go ignored for ages. Those of us who are old enough remember an era where a huge number of web sites only targeted Trident and didn't care in the slightest whether it worked in any other engine.

There has to be an opposite to Postel's Law that acknowledges it's better in some cases to ensure that breakage for anybody becomes breakage for everybody because that means the breakage can't be ignored.

If authored files had to be valid in order to work, how would the author have sold you an invalid file in the first place? They would have seen that it didn’t work when they were making it, and fixed it. If they’d sold you a book that didn’t open, you’d be entitled to a refund.

There is an evolutionary counter-argument here.

Imagine two file formats competing to be "the web":

* In file format A, the syntax is narrowly specified and any irregularity causes it to be not rendered at all.

* In file format B, there are many ways to express the same thing and many syntactic forms are optional and can be omitted while leaving a page that still renders.

Now imagine millions of non-technical users are trying to make websites. They randomly pick A or B. Which ones do you think are more likely to get to a point where they have a page that renders that they can put online? It's B.

Even though the file format of B offends our software engineer sensibilities, in an evolutionary sense it is much more robust and able to thrive because more fallible users are able to successfully write working files using it. It will completely dominate once network effects get involved.

1. ensures that 2. doesn't happen. Not doing 1. pretty much gurantees that 2. will happen.