Module encoding/html

HTML entity decoding — named, decimal, and hexadecimal character references.

Decodes named (&), decimal (&), and hex (&) HTML character references. Uses Legacy mode — entities without trailing semicolon are also decoded.

Example

html :: import("std/encoding/html");

result := html.decode(`& < & &`);
assert((result == `& < & &`), "decoded entities");

Stability

unstable — html_decode takes one argument, and the WHATWG spec needs two. Its §13.2.5.72 named-reference rule is CONTEXT-dependent: the semicolon-less legacy form (&amp for &) is decoded in TEXT but must stay literal in an ATTRIBUTE VALUE when the next character is = or alphanumeric, so ?a=1&lang=x keeps its &lang instead of becoming &lang the entity. This module decodes legacy references unconditionally, because fn(input : String) -> String has nowhere to say which position the text came from. Fixing it means a second parameter or a second entry point, so it is not additive, and it is mildly security-relevant in the direction of mangling URLs rather than of injecting markup.

html_encode's set is settled — the five XSS-critical characters, chosen so html_decode(html_encode(s)) == s holds for any s — and that round-trip is what a fix to the above must not break.

One smaller thing to tidy before freezing: this module RE-EXPORTS is_valid_entity_code and from_code_point, which html_char_utils also exports, so two module paths publish the same two names. Whichever becomes canonical, the other's export goes — a removal, not an addition.

The entity tables are no longer a hazard: they are built once at module init (_entity_map, _legacy_set), where they used to be lazily built through an unsynchronised flag that two threads' first decodes raced on (issues/fixed/html-and-log-globals-raced.md), and html_decode appends in place instead of rebuilding the result through a template string per character, which was O(n²).