getAttribute("class") trims the value and returns "" when absent #342

Activity

1. osazemeu commented on Oct 26, 2026

   

   osazemeu

   on Oct 26, 2026

   More actions

   I can reproduce both halves of this on 0.18.14 and on main at e2ee208, and I can open a PR. Which fix would you take?

   1. Only the null half. getAttribute('class') returns null when the element has no class attribute. This is one line in Element#getAttribute. We ship it as a local patch today, because a parser that falls back to a default only on null (TipTap's Link mark, for one) reads '' as "class set to empty".
   2. Both halves. The class attribute keeps the value as written, and the token list derives from it. Today the parser (parse-from-string.js), Element#setAttribute and the setAttribute helper in shared/attributes.js all write through className, which splits the value and joins it again. So the stored Attr value is already normalized, and the selector path, which since perf: read class attribute directly in selector adapter #339 reads getAttributeNode('class')?.value ?? null, also sees "ab". A fix would store the raw value and rewrite it only when classList itself is mutated. That overlaps the parse path perf: skip the token list while parsing class attributes #336 changes, so I would rebase on it, or wait for it, whichever you prefer.

   Repro on 0.18.14 (unpatched) and on main at e2ee208 (Node 24.14.1), extended with an explicit empty class:

   import { parseHTML } from "linkedom";
   const { document } = parseHTML('<html><body><i class=" ab"></i><b></b><u class=""></u></body></html>');
   const [i, b, u] = document.body.children;
   i.getAttribute("class");                        // "ab"   (expected " ab")
   b.getAttribute("class");                        // ""     (expected null)
   u.getAttribute("class");                        // ""     (correct)
   b.hasAttribute("class");                        // false  (correct)
   document.querySelectorAll('[class^=" a"]').length; // 0   (expected 1)