[dashorst]

Pages now render lang and dir on their <html> tag from the page locale: dir="rtl" when that language is written right to left, dir="ltr" otherwise. Screen readers pick their pronunciation from lang (WCAG 2.2 SC 3.1.1, level A). Until now an application had to bind <html> to a TransparentWebMarkupContainer to get it.

WebPage#onHtmlTag(ComponentTag) receives a mutable copy of the <html> tag on every full render:

• lang: set to getLocale().toLanguageTag(), unless the markup has lang or xml:lang, or the locale has no language (Locale.ROOT).
• dir: set to rtl when Session.isRtlLanguage (WICKET-6943) recognises the language and to ltr otherwise, unless the markup has its own dir or declares its own lang. Writing ltr explicitly follows solomax's review and MDN's recommendation; 4f19938 explains why.
• Opting out: put dir="ltr" on <html> in the markup, or call tag.remove("dir") after super.onHtmlTag(tag) in the base page.
• Extending: override the method to add document-wide attributes, such as a theme class or data-theme.

The commit messages carry the design decisions: why there is no setting, no IOnComponentTagListener call, no contributor mechanism and no hook for <head> or <body>. It also explains why <html> attributes can't change on an Ajax request. The draft migration-guide wording is there too.

Where to look

• Core:

  • WebPage: renderNext and onHtmlTag.
  • HtmlTagHandler (new, page markup only): keeps each <html> tag as a tag with id _html_ and the new ComponentTag.RENDERED_BY_PAGE flag. No component is created.
  • MarkupContainer and TagUtils: queueing ignores the flagged tag, and findTagMarkup keeps looking inside it, as it does for <head>. Without this, components inside <html> are no longer found or dequeued.
  • HtmlTagTest: 21 tests covering locale changes between renders, markup lang/xml:lang/dir in any case, Locale.ROOT, RTL and the Latin-script exception, the opt-out, a bound <html wicket:id>, markup inheritance, queueing and an Ajax update.

• Mechanical:

  • 235 expected-output files in wicket-core-tests and wicket-extensions, each changed only on its <html> line (lang and dir added).
  • Two Java tests with a hard-coded <html>.

• Small test fixes:

  • AjaxBehaviorEnabledTest built its own WicketTester and rendered with the JVM default locale, so it failed on non-US machines. It now uses newWicketTester, which pins Locale.US.
  • The OpenCloseTagExpander test markup now quotes an xmlns:wicket value, because the parser mangles unquoted values containing / or : (Unquoted attribute values containing / or : are cut off by the markup parser #1864).

• Migration guide: migration-to-12.md, now on master, gets a section on the change. The branch was merged with master to resolve the conflict on that file; the merge keeps both sections.

Risk

• Every page's output changes. Pages without lang in their markup get lang and dir. Tests that compare rendered pages with expected files need updating.
• The <html> tag is now written from its parsed attributes, like a component tag. Values get double quotes, a valueless xmlns:wicket becomes xmlns:wicket="", and a tag spread over several lines is written on one.
• RTL layout changes. An application with RTL translations and an LTR-only stylesheet sees its layout mirror until it opts out.

Verification

• mvn clean verify -Pjs-test passed in two parts: the first run stopped at one extensions expected file, and the run resumed from wicket-extensions passed after fixing it.
• CI passes on every pushed commit. With explicit dir, wicket-core-tests (2571) and wicket-extensions (141) pass locally.
• wicket-core-tests and wicket-extensions also pass with the JVM locale set to nl_NL.

Closes #1848

Generated-by: Claude Code (Claude Opus 5.5)

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 62.75%. Comparing base (7d4b720) to head (fab7cc7).
⚠️ Report is 6 commits behind head on master.

Additional details and impacted files

@@             Coverage Diff              @@
##             master    #1870      +/-   ##
============================================
+ Coverage     62.72%   62.75%   +0.02%     
- Complexity    11512    11534      +22     
============================================
  Files          1254     1255       +1     
  Lines         48665    48698      +33     
  Branches       6845     6855      +10     
============================================
+ Hits          30526    30559      +33     
  Misses        15463    15463              
  Partials       2676     2676              

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[papegaaij]

This will go in with a merge commit, per AGENTS.md; 4f19938 already explains the switch to writing dir="ltr", so the history reads fine without squashing.

[papegaaij]

HTML attribute names are case-insensitive, but getAttribute is a plain map lookup. <html LANG="en" DIR="ltr"> doesn't count as having lang, and renders as <html LANG="en" DIR="ltr" lang="en-US" dir="ltr">. Browsers take the first one, but validators and accessibility checkers such as axe flag the duplicates. The same applies to XML:LANG and to the dir check below.

ValueMap.getKey(String) already does a case-insensitive lookup:

ValueMap attributes = tag.getAttributes();
if (attributes.getKey("lang") != null || attributes.getKey("xml:lang") != null)
{
	return;
}
...
if (attributes.getKey("dir") == null)

A case in HtmlTagTest with upper-case attributes would cover it.

[dashorst]

Good catch, thanks. fab7cc7 checks lang, xml:lang and dir with getKey now, as you suggested. HtmlTagTest has three new cases with upper-case attributes: LANG/DIR, XML:LANG and DIR alone. Before the fix they produced exactly the duplicates you described.