#5605 fix Chinese search index (#5611)

generate search index for Latin words correctly if search language is Chinese

4 files changed, 39 insertions, 2 deletions

diff --git a/CHANGES b/CHANGES
index 867cdc950..d3f8b0a0e 100644
--- a/CHANGES
+++ b/CHANGES
@@ -15,6 +15,8 @@ Features added
 
 Bugs fixed
 ----------
+* #5605 If the documentation language is set to Chinese, English words could not
+  be searched.
 
 Testing
 --------
diff --git a/sphinx/search/zh.py b/sphinx/search/zh.py
index 6c5b65d6b..3753bc990 100644
--- a/sphinx/search/zh.py
+++ b/sphinx/search/zh.py
@@ -233,7 +233,8 @@ class SearchChinese(SearchLanguage):
     language_name = 'Chinese'
     js_stemmer_code = js_porter_stemmer
     stopwords = english_stopwords
-    latin1_letters = re.compile(u'(?u)\\w+[\u0000-\u00ff]')
+    latin1_letters = re.compile(r'[a-zA-Z0-9_]+')
+    latin_terms = []  # type: List[unicode]
 
     def init(self, options):
         # type: (Dict) -> None
@@ -250,7 +251,9 @@ class SearchChinese(SearchLanguage):
         if JIEBA:
             chinese = list(jieba.cut_for_search(input))
 
-        latin1 = self.latin1_letters.findall(input)
+        latin1 = \
+            [term.strip() for term in self.latin1_letters.findall(input)]
+        self.latin_terms.extend(latin1)
         return chinese + latin1
 
     def word_filter(self, stemmed_word):
@@ -259,4 +262,15 @@ class SearchChinese(SearchLanguage):
 
     def stem(self, word):
         # type: (unicode) -> unicode
+
+        # Don't stem Latin words that are long enough to be relevant for search
+        # if not stemmed, but would be too short after being stemmed
+        # avoids some issues with acronyms
+        should_not_be_stemmed = (
+            word in self.latin_terms and
+            len(word) >= 3 and
+            len(self.stemmer.stem(word.lower())) < 3
+        )
+        if should_not_be_stemmed:
+            return word.lower()
         return self.stemmer.stem(word.lower())
diff --git a/tests/roots/test-search/tocitem.rst b/tests/roots/test-search/tocitem.rst
index f082abf65..5d99f0a66 100644
--- a/tests/roots/test-search/tocitem.rst
+++ b/tests/roots/test-search/tocitem.rst
@@ -8,3 +8,9 @@ textinheading
 =============
 
 lorem ipsum
+
+可以查看 FAQ 模块中 Chinesetest 部分
+
+模块中 CAS service部分
+
+可以Chinesetesttwo查看
diff --git a/tests/test_search.py b/tests/test_search.py
index 886151831..4c7eb8b21 100644
--- a/tests/test_search.py
+++ b/tests/test_search.py
@@ -240,3 +240,18 @@ def test_IndexBuilder_lookup():
     # zh_CN
     index = IndexBuilder(env, 'zh_CN', {}, None)
     assert index.lang.lang == 'zh'
+
+
+@pytest.mark.sphinx(
+    testroot='search',
+    confoverrides={'html_search_language': 'zh'},
+    srcdir='search_zh'
+)
+def test_search_index_gen_zh(app, status, warning):
+    app.builder.build_all()
+    # jsdump fails if search language is 'zh'; hence we just get the text:
+    searchindex = (app.outdir / 'searchindex.js').text()
+    assert 'chinesetest ' not in searchindex
+    assert 'chinesetest' in searchindex
+    assert 'chinesetesttwo' in searchindex
+    assert 'cas' in searchindex