The tag omission/inference rules for <p> come from SGML, and, as presented in WHATWG's HTML spec, seem arbitrary. But actually what they're saying is that a par

The tag omission/inference rules for <p> come from SGML, and, as presented in WHATWG's HTML spec, seem arbitrary. But actually what they're saying is that a paragraph is automatically ended on a non-phrasing element. In earlier HTML versions, a definition of flow content was given like this: <!ENTITY % flow "%block;|%inline"> where the flow, block, and inline parameter entities expand to strings like a|abbr|... such that they can be used in element declarations as follows: <!ELEMENT p (#PCDATA|%phrasing -(%flow_only|figcaption)> meaning the content of p can be any phrasing element (and text content), except elements that are in the flow_only category or the figcaption element, capturing precisely HTML5's rules. While Ian Hickson's initial (HTML5.0) rules were exactly corresponding to those of HTML4.01, later HTML5 revisions (^2) indeed failed to update the prose description for p-terminating elements, for the very reason that those weren't presented as a formal spec (such as a DTD). In a presentation I held [1] I made exactly that point for the p element, and you might find the derivation of a new SGML DTD grammar for HTML5.x generally of interest. [1]: http://sgmljs.net/docs/html5-dtd-slides-wrapper.html http://sgmljs.net/docs/html5-dtd-slides-wrapper.html ^2: if WHATWG had bothered to version their specs, and W3C still would produce versioned snapshots of those, at least