Sort MO indexes by encoded context and message keys

babel/messages/mofile.py

@@ -196,6 +196,8 @@ def write_mo(fileobj: SupportsWrite[bytes], catalog: Catalog, use_fuzzy: bool =
     # Each entry has first the size of the string, then the file offset.
     koffsets = []
     voffsets = []
+    # 2026-10-10: MO lookup order uses encoded contexts and singular IDs.
+    offsets.sort(key=lambda item: ids[item[0] : item[0] + item[1]].split(b"\x00", 1)[0])
     for o1, l1, o2, l2 in offsets:
         koffsets += [l1, o1 + keystart]
         voffsets += [l2, o2 + valuestart]

tests/messages/test_mofile.py

@@ -104,3 +104,37 @@ def test_read_mo_decodes_message_context() -> None:
     assert message.context == 'ほげ'
     assert message.string == 'åäöÅÄÖ'
     assert not catalog.get('x')  # Not without the context
+
+
+def test_context_originals_are_sorted():
+    # 2026-10-10: GNU MO lookup tables sort the encoded original keys.
+    import struct
+
+    for charset in ['utf-8', 'cp1252']:
+        catalog = Catalog(charset=charset)
+        catalog.add('a', 'last', context='z')
+        catalog.add('z', 'first', context='a')
+        catalog.add(('b', 'bs'), ('one', 'many'), context='y')
+        catalog.add('plain', 'plain translation')
+        catalog.add('c', 'euro', context='\u20ac')
+        catalog.add('d', 'accent', context='\u00e9')
+        buf = BytesIO()
+        mofile.write_mo(buf, catalog)
+        data = buf.getvalue()
+        magic = struct.unpack_from('<I', data)[0]
+        order = '<' if magic == mofile.LE_MAGIC else '>'
+        _, _, count, originals, translations = struct.unpack_from(order + '5I', data)
+        pairs = []
+        for index in range(count):
+            length, offset = struct.unpack_from(order + '2I', data, originals + index * 8)
+            key = data[offset : offset + length].split(b'\x00', 1)[0]
+            length, offset = struct.unpack_from(order + '2I', data, translations + index * 8)
+            pairs.append((key, data[offset : offset + length]))
+        assert [key for key, _ in pairs] == sorted(key for key, _ in pairs)
+        entries = dict(pairs)
+        assert entries[b'a\x04z'] == b'first'
+        assert entries[b'z\x04a'] == b'last'
+        assert entries[b'y\x04b'] == b'one\x00many'
+        assert entries[b'plain'] == b'plain translation'
+        assert entries['\u20ac\x04c'.encode(charset)] == b'euro'
+        assert entries['\u00e9\x04d'.encode(charset)] == b'accent'