ensure lists in the search index are also deterministic

tests/test_search.py

@@ -308,19 +308,33 @@ def test_parallel(app):
 
 @pytest.mark.sphinx(testroot='search')
 def test_search_index_is_deterministic(app):
-    app.builder.build_all()
-    index = load_searchindex(app.outdir / 'searchindex.js')
-
-    def check_if_dicts_are_sorted(item):
+    LISTS_NOT_TO_SORT = [
+        # Elements of .titles correspond to the element of .docnames in the same position.
+        ".titles",
+        # Elements of .filenames correspond to the element of .docnames in the same position.
+        ".filenames",
+    ]
+
+    # In the search index, titles inside .alltitles are stored as a tuple of
+    # (document_idx, title_anchor). Tuples are represented as lists in JSON,
+    # but their contents must not be sorted. We cannot sort them anyway, as
+    # document_idx is an int and title_anchor is a str.
+    def is_title_tuple_type(item):
+        return len(item) == 2 and isinstance(item[0], int) and isinstance(item[1], str)
+
+    def assert_is_sorted(item, path):
+        print('visiting', path if path else '<root>')
         if isinstance(item, dict):
             assert list(item.keys()) == sorted(item.keys())
-            for child in item.values():
-                check_if_dicts_are_sorted(child)
-        # Lists in the search index cannot be sorted: for some lists, their
-        # position inside the list is referenced elsewhere in the index, so if
-        # we were to sort lists, the search index would break.
+            for key, value in item.items():
+                assert_is_sorted(value, f'{path}.{key}')
         elif isinstance(item, list):
-            for child in item:
-                check_if_dicts_are_sorted(child)
+            if not is_title_tuple_type(item) and path not in LISTS_NOT_TO_SORT:
+                assert item == sorted(item)
+            for i, child in enumerate(item):
+                assert_is_sorted(child, f'{path}[{i}]')
 
-    check_if_dicts_are_sorted(index)
+    app.builder.build_all()
+    index = load_searchindex(app.outdir / 'searchindex.js')
+    print(f'index contents:\n{json.dumps(index, indent=2)}\n')  # Pretty print the index.
+    assert_is_sorted(index, '')