[RFC] Decoding HTML and the Ambiguous Ampersand

6 months ago by Dennis Snell — view source

unread

Greetings all,

The html_entity_decode( … ENT_HTML5 … ) function has a number of issues that I’d like to correct.

It’s missing 720 of HTML5’s specified named character references.

106 of these are named character references which do not require a trailing semicolon, such as &acute

It’s unaware of the ambiguous ampersand rule, which allows these 106 in special circumstances.

HTML5 asserts that the list of named character references will not expand in the future. It can be found authoritatively at the following URL:

https://html.spec.whatwg.org/entities.json https://html.spec.whatwg.org/entities.json

The ambiguous ampersand rule smoothes over legacy behavior from before HTML5 where ampersands were not properly encoded in attribute values, specifically in URL values. For example, in a query string for a search, one might find ?q=dog&not=cat. The &not in that value would decode to U+AC (¬), but since it’s in an attribute value it will be left as plaintext. Inside normal HTML markup it would transform into ?q=dog¬=cat. There are related nuances when numeric character references are found at the end of a string or boundary without the semicolon.

The function signature of html_entity_decode() does not currently allow for correcting this behavior. I’d like to propose an RFC or a bug fix which either extends the function (perhaps by adding a new flag like ENT_AMBIGUOUS_AMPERSAND) or preferably creates a new function. For the missing character references I wonder if it would be enough to add them to the list of default translatable references.

One challenge with the existing function is that the concept of the translation table stands in contrast with the fixed and static nature of HTML5’s replacement tables. A new function or set of functions could open up spec-compliant decoding while providing helpful methods that are necessary in many common server-side operations:

html_decode( ‘attribute’ | ‘data’, $raw_text, $input_encoding = ‘utf-8' )

html_text_contains( ‘attribute’ | ‘data’, $raw_haystack, $needle, $input_encoding = ‘utf-8’ )

html_text_starts_with( ‘attribute’ | ‘data’, $raw_haystack, $needle, $input_encoding = ‘utf-8’ )

These methods are handy for inspecting things like encoded attribute values in a memory-efficient and processing-efficient way, when it’s not necessary to decode the entire value. In common situations, one encounters data-URIs with potentially megabytes of image data and processing only the first few or tens of bytes can save a lot of overhead.

We’re exploring pure-PHP solutions to these problems in WordPress in attempts to improve the reliability and safety of handling HTML. I’d love to hear your thoughts and know if anyone is willing to work with me to create an RFC or directly propose patches. We’ve created a step function which allows finding the next character reference and decoding it separately, enabling some novel features like highlighting the character references in source text.

Should I propose an RFC for this?

Warmly,
Dennis Snell
Automattic Inc.

Thanks everyone for your feedback so far on the decode_html() RFC [https://wiki.php.net/rfc/decode_html https://wiki.php.net/rfc/decode_html]

I’ve updated it replacing the new constants with a new HtmlContext enum, and the interface seems much nicer this way. I particularly like how PHP enforces passing a valid value, vs. hoping that the right flag is used.

Additionally I added a section that I previously forgot, which highlights the source of the infamous mojibake/gremlins. HTML has special rules for remapping the C1 control characters, as if they had been stored or recorded for Windows-1251.

Warmly,
Dennis Snell

Hi Dennis

+1 on the concept.
I just have two concerns:

Thanks Niels. I appreciate the help you’ve already provided on this process, and the work you’ve done with lexbor.

I'm not so sure that the name "decode_html" is self-descriptive enough, it sounds very generic.

The name is not very important to me. For the sake of history, the reason I have chosen “decode HTML” is because, unlike an HTML parser, this is focused on taking a snippet of HTML “text” content and decoding it into a “plain PHP string.”

The existing html_entity_decode() is very close in naming but ties this concept into entities, and overlooks other basic text decoding concerns (newline normalization and NULL byte handling).

Originally I had “utf8” in the name but someone else thought it was too long and specific. I want the name to educate developers and also be terse. Naming is hard.

I would strongly suggest to explore an implementation based on Lexbor. I'm pretty confident that it can be done by reusing the internal APIs. The advantage is that it will be less code to maintain. You pull off some fancy tricks in your implementation for performance reasons, but that also adds to complexity and maintenance burden. Also since this is C, we must be extra careful when implementing tricks.

Yeah I agree and I’ll share more below. The tricks I’m using in my PR implementing the RFC are partly there to propose adoption into PHP and partly there to get a real sense of my algorithm vs. those found in Chrome, Firefox, Safari, and lexbor. I’ve attempted to build a search algorithm for named character references that optimizes for cache locality in contrast to algorithmic complexity where RAM access is assumed to be free.

My code isn’t currently well document and doesn’t meet the PHP-src coding standards, but the algorithm is pretty basic and easy to explain. It’s also “unoptimized” for C, mostly. I think there are still large gains to be made that so far I’ve been unable to visualize incorporating into the lexbor parser. For example, decode_html() assumes we’re starting already with a span of text that is HTML text. We’re not making conditional decisions on whether the next byte produces a token that escapes out of the text parsing mode.

If we could have a single implementation, that would be great. I do understand of course your concern that DOM is not a required extension, and therefore basing the internals on Lexbor makes it tied to the DOM extension which may not be available. I however suspect that a large chunk of people needing a function like this have DOM available (as DOM is required by many HTML-processing-related packages). I can also look into it sometime soon if you want; anyway feel free to ping me.

I’m also very open to lexbor-based approaches but I’ve so-far found it more complicated than I expected. In some part this is because it involves setting up the parser and state machine for the HTML specification and much of the actual decoding can be safely done without this.

The other part is the extension aspect. I hear you, that you would expect calling code to have the DOM extensions available, but that’s simply not the case when developing a platform like WordPress, which I do. We don’t have control over the servers or environments where people are deploying this, and the availability of the DOM extensions is low enough that WordPress code simply cannot use DOMDocument (even though it shouldn’t because of the wild problems that has for attempting to parse HTML).

People resort to html_entity_decode() because that’s the only option. In WordPress we now have a spec-compliant decoder, but as it’s in user-space PHP its performance is far below what’s possible.

I’d love your help in setting up lexbor’s state machine to decode text nodes. I’d love it even more if this could be part of the PHP language. It constantly surprises me that the language of the web (PHP) doesn’t have the tools to speak the language of the web (HTML). This RFC is all about taking a step towards ensuring that PHP developers can rely on PHP to be a reliable middle-man between the HTML domain and the PHP domain.

In other words, requiring the DOM extension or DOM\HtmlDocument would be such a non-starter for WordPress (accounting for 43% of the web today) that it would completely unavailable.

And I do have the following thoughts:

We should amend the ENT_HTML5 related docs already that it's not compliant.

Perhaps ENT_HTML5 should be deprecated. E.g. you could say in your RFC that ENT_HTML5 will be deprecated in the release after the version that will have decode_html(). The reason I suggest the release after and not the same release is because I strongly believe that we should have at least one version where the proper alternative is available without forcing a deprecation to users already.

I love this suggestion. Just for reference, since I’ve looked before and not found it. Can someone indicate where to find the PHP function documentation? There are a number of updates I would love to propose but I don’t know where to find the content that appears in https://www.php.net/manual/en/function.html-entity-decode.php, for instance.

Kind regards
Niels

Mad respect to the work you’ve brought to lexbor and to PHP. I’m excited to start relying on \DOM\HtmlDocument and have started using it in my benchmarks and HTML analysis as we develop the WordPress HTML API (a streaming, low memory-overhead, reentrant HTML parsing and manipulation framework in user-space PHP).

Dennis Snell