Based on the discussion with Just and anthrotype: Undo the changes in the name table and extend unittests. If there are inconsistencies in the name table, it's ok to create new name IDs, even if this might not be the most efficient way of creating the name table.

Lib/fontTools/ttLib/tables/_n_a_m_e.py

@@ -311,16 +311,6 @@ class table__n_a_m_e(DefaultTable.DefaultTable):
 			if nameID is not None:
 				return nameID
 			nameID = self._findUnusedNameID()
-
-			nameID_win = self.findMultilingualName(names, windows=True, mac=False)
-			nameID_mac = self.findMultilingualName(names, windows=False, mac=True)
-			if nameID_win and nameID_mac is None:
-				nameID = nameID_win
-				windows = False
-			if nameID_mac and nameID_win is None:
-				nameID = nameID_mac
-				mac = False
-
 		# TODO: Should minimize BCP 47 language codes.
 		# https://github.com/fonttools/fonttools/issues/930
 		for lang, name in sorted(names.items()):

Tests/otlLib/builder_test.py

@@ -1462,15 +1462,16 @@ def test_buildStatTable_name_duplicates():
 
     builder.buildStatTable(font_obj, AXES, windowsNames=True, macNames=True)
     actual_names = [x.string for x in font_obj["name"].names]
-    expected_names = ['Weight', 'Weight',
-                      'ExtraLight', 'ExtraLight',
-                      'Light', 'Light',
-                      'Regular', 'Regular',
-                      'Medium', 'Medium',
-                      'Bold', 'Bold',
-                      'ExtraBold', 'ExtraBold',
-                      'Black', 'Black']
-    # Multiple name records were added by buildStatTable
+    expected_names = ['Weight', 'Weight', 'Weight',
+                      'ExtraLight', 'ExtraLight', 'ExtraLight',
+                      'Light', 'Light', 'Light',
+                      'Regular', 'Regular', 'Regular',
+                      'Medium', 'Medium', 'Medium',
+                      'Bold', 'Bold', 'Bold',
+                      'ExtraBold', 'ExtraBold', 'ExtraBold',
+                      'Black', 'Black', 'Black']
+    # Because there is an inconsistency in the names add new name IDs
+    # for each platform -> windowsNames=True, macNames=True
     assert sorted(expected_names) == sorted(actual_names)
 
 

Tests/ttLib/tables/_n_a_m_e_test.py

@@ -317,11 +317,24 @@ class NameTableTest(unittest.TestCase):
 		font_obj["name"].names = []
 
 		font_obj["name"].setName('Weight', 270, 3, 1, 0x409)
+		font_obj["name"].setName('Weight', 270, 1, 0, 0)
 		names = {'en': 'Weight', }
 		nameID = font_obj["name"].addMultilingualName(names, minNameID=256)
-
 		self.assertEqual(270, nameID)
 
+		font_obj["name"].removeNames(nameID=270, platformID=1)  # remove Mac name
+		nameID = font_obj["name"].addMultilingualName(names, minNameID=256)
+		# Because there is an inconsistency in the names add a new name ID
+		self.assertEqual(271, nameID)
+
+		font_obj["name"].removeNames(platformID=3)  # remove all Windows names
+		font_obj["name"].removeNames(platformID=1)  # remove all Mac names
+		nameID = font_obj["name"].addMultilingualName(names, minNameID=256)
+		#Because there is no name ID with the name 'Weight',
+		# take the next available name ID -> minNameID=256
+		self.assertEqual(256, nameID)
+
+
 
 	def test_decompile_badOffset(self):
                 # https://github.com/fonttools/fonttools/issues/525