Derive JS class name from stemmer filename to fix Chinese search (#14416)

Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>

CHANGES.rst

@@ -8,6 +8,11 @@ Bugs fixed
   styled with ``'colorrows'`` (which is the default).
   Patch by Jean-François B.
 
+* #14229: Fix JavaScript search for languages whose stemmer class name
+  does not match the language name, such as Chinese (which reuses the
+  English stemmer) and Dutch (which uses the Dutch Porter stemmer).
+  Patch by Hugo van Kemenade
+
 
 Release 9.1.0 (released Dec 31, 2025)
 =====================================

sphinx/search/__init__.py

@@ -582,10 +582,21 @@ class IndexBuilder:
 
         base_js_path = _MINIFIED_JS_PATH / 'base-stemmer.js'
         language_js_path = _MINIFIED_JS_PATH / self.lang.js_stemmer_rawcode
+        # Derive the JS class name from the stemmer filename rather than
+        # from language_name, since some languages reuse another language's
+        # stemmer. For example, SearchChinese reuses english-stemmer.js,
+        # which defines EnglishStemmer.
+        stemmer_class = (
+            self.lang.js_stemmer_rawcode.removesuffix('-stemmer.js')
+            .title()
+            .replace('_', '')
+            .replace('-', '')
+            + 'Stemmer'
+        )
         return '\n'.join((
             base_js_path.read_text(encoding='utf-8'),
             language_js_path.read_text(encoding='utf-8'),
-            f'window.Stemmer = {self.lang.language_name}Stemmer;',
+            f'window.Stemmer = {stemmer_class};',
         ))
 
 

tests/test_search.py

@@ -393,6 +393,32 @@ def test_search_index_gen_zh(app: SphinxTestApp) -> None:
     assert 'chinesetesttwo' in index['terms']
     assert 'cas' in index['terms']
 
+    language_data = (app.outdir / '_static' / 'language_data.js').read_text(
+        encoding='utf-8'
+    )
+    # SearchChinese reuses english-stemmer.js
+    assert 'var EnglishStemmer' in language_data
+    assert 'window.Stemmer = EnglishStemmer;' in language_data
+    assert 'ChineseStemmer' not in language_data
+
+
+@pytest.mark.parametrize(
+    ('js_stemmer_rawcode', 'stemmer_class'),
+    [
+        ('english-stemmer.js', 'EnglishStemmer'),
+        # dutch_porter-stemmer.js is the only one with an underscore
+        ('dutch_porter-stemmer.js', 'DutchPorterStemmer'),
+    ],
+)
+def test_js_stemmer_class_name(js_stemmer_rawcode: str, stemmer_class: str) -> None:
+    # check that the JS class name is correctly derived from the filename
+    env = DummyEnvironment('1.0', DummyDomainsContainer())
+    index = IndexBuilder(env, 'en', {}, '')  # type: ignore[arg-type]
+    index.lang.js_stemmer_rawcode = js_stemmer_rawcode
+    js = index.get_js_stemmer_code()
+    assert f'var {stemmer_class}' in js
+    assert f'window.Stemmer = {stemmer_class};' in js
+
 
 @pytest.mark.sphinx(
     'html',