Inline HTML

UJO (Universal JSON Output) keeps inline formatting inside its texts as HTML: Hello <b>world</b>. A block editor's own sanitizer runs in the writer's browser, which a request can skip; phil/ujo sanitizes every HTML value again on the server, field by field, when a document is normalized and again when it is rendered.

The sanitizer

InlineSanitizer::sanitize($html, $profile) reads the fragment token by token — text, start tags with their attributes, end tags, comments — and writes back only what the profile allows, in one canonical form:

• text is decoded, then escaped again: é becomes é, < becomes <;
• an element no tool of the profile writes is unwrapped: its text stays;
• script, style, iframe, object, embed, template, noscript, textarea, select, svg, math, title, head, frameset, frame, applet, audio, video, canvas, form and annotation-xml are dropped with their content, unless the profile reads them: rich and mathml read math;
• a tool kept only within other elements is unwrapped elsewhere: a ruby's <rt> outside a <ruby>, an element of MathML outside a <math>;
• comments are dropped;
• attributes a tool does not name are dropped, those it names are cleaned, and an element missing a required one is unwrapped;
• elements are closed in order: a stray end tag is ignored, an element left open is closed, and a link is never nested in a link.

InlineSanitizer::toText($html) gives the decoded text without elements, a <br> as a new line.

Profiles

Inline tools of rich

Free colours — <font color>, style="color:…" — are unwrapped: a colour is a token the look defines.

Adding a profile or a tool

$profile = InlineProfile::rich()
    ->without('marker')
    ->with(new InlineTool('keyboard', 'kbd', ['kbd']));

$sanitizer = new InlineSanitizer(['rich' => $profile, 'caption' => InlineProfile::plain()]);

An InlineTool has a name, the tag it writes, the patterns it reads — span, span.class, a[attribute] —, an optional class it writes, the attributes it keeps with an AttributeValue each — Link, Media, Token, Reference, Text, Target, Rel, Length, Number —, and the attributes it requires. A profile tries its tools in order: a narrower pattern comes before a wider one. Give the sanitizer to both Normalizer and HtmlRenderer.

Why its own sanitizer

phil/markup purifies whole documents at a level — basic, forum, all — through symfony/html-sanitizer. UJO needs less and more: a rule per field rather than per document, the classes and data attributes of the Editor.js inline tools kept exactly, and a canonical form so that a stored document compares equal to itself. And the readers of phil/markup convert other formats into UJO, so phil/ujo cannot depend on it. An application that wants a second pass may give the rendered HTML to phil/markup's purifier at the all level; with the default profiles nothing is removed by it.