accname

Accessible Name and Description Computation per WAI-ARIA (accname 1.2) and HTML-AAM, generic over a11y-dom. The entry point for everything that needs a name.

Status

0.26.0 on crates.io. Checked differentially against Chrome: auditmysite has accname-diff for this; on the first corpus 381 of 395 names matched, and the deviations all came from contents — the dominant cause is CSS text-transform, which Chrome applies to the name and accname does not.

Structure

Public surface

Entry	Purpose
name, description	the two computations
role, implicit, allows_name_from_content, name_is_prohibited	role knowledge the step sequence needs
IdIndex	resolution of aria-labelledby/aria-describedby without repeated tree searches

Conformance corpus

The web-platform-tests cases accname/name/*.html are vendored under crates/accname/tests/corpus/wpt (xtask/vendor-wpt-accname.sh, BSD-3-Clause, origin in UPSTREAM.md) and run by tests/wpt.rs: every element with data-expectedlabel is one case. Failures from documented gaps (::before/::after, CSS counters, text-transform, <style> blocks, values set by script, shadow DOM built by script) are listed in tests/wpt_known_failures.txt; the test fails both for an unlisted failure and for a listed case that now passes (WPT_BLESS=1 regenerates the list).

Dependencies

Downward: a11y-dom. Upward: a11y-rules, hosts with document-based collection.

Limits

name() sees no CSS: text-transform, ::before/::after and the browser’s visibility rules are not modeled. A space therefore always separates the parts of a name, even for inline elements: <abbr>EU</abbr>-Arktis yields “EU -Arktis”, Chrome “EU-Arktis”. Only what carries hidden or aria-hidden="true" counts as hidden. A decision based on the tag was tried in 0.11.1 and reverted in 0.11.2 — on real pages <span> is often a block element via CSS (auditmysite corpus: 450 new deviations against 5 fixed).

Hosts with computed styles (Rendering, tier 3) call name_rendered(doc, node, ids). There, children with display: inline or contents attach directly, all others and replaced elements (img, svg, …) are set apart, <br> always separates; display: none (also on an ancestor) and visibility: hidden hide content — a <label> hidden this way also no longer names its field — except for an explicit reference via aria-labelledby. The host must supply the whitespace text nodes — CDP DOM.getDocument omits them, DOMSnapshot does not. In the auditmysite corpus (35 pages) this leaves one real deviation pattern instead of six. text-transform and ::before/::after remain out of scope there as well.

<dt> (role term) gets no name from content: term is listed neither in ARIA 1.2 nor in the ARIA 1.3 draft under “name from content”, and in the draft explicitly under “name prohibited”. Chrome nevertheless names <dt> from its content — an expected deviation. Where a host has native values (Chrome via CDP), these are to be marked separately — not silently swapped for this package’s own computation.