use clean_underscore fixture

spacy/tests/doc/test_underscore.py

@@ -7,6 +7,15 @@ from spacy.tokens import Doc, Span, Token
 from spacy.tokens.underscore import Underscore
 
 
+@pytest.fixture(scope="function", autouse=True)
+def clean_underscore():
+    # reset the Underscore object after the test, to avoid having state copied across tests
+    yield
+    Underscore.doc_extensions = {}
+    Underscore.span_extensions = {}
+    Underscore.token_extensions = {}
+
+
 def test_create_doc_underscore():
     doc = Mock()
     doc.doc = doc

spacy/tests/matcher/test_matcher_api.py

@@ -6,6 +6,7 @@ import re
 from mock import Mock
 from spacy.matcher import Matcher, DependencyMatcher
 from spacy.tokens import Doc, Token
+from ..doc.test_underscore import clean_underscore
 
 
 @pytest.fixture
@@ -200,6 +201,7 @@ def test_matcher_any_token_operator(en_vocab):
     assert matches[2] == "test hello world"
 
 
+@pytest.mark.usefixtures("clean_underscore")
 def test_matcher_extension_attribute(en_vocab):
     matcher = Matcher(en_vocab)
     get_is_fruit = lambda token: token.text in ("apple", "banana")

spacy/tests/regression/test_issue4849.py

@@ -9,11 +9,6 @@ from spacy.tokens.underscore import Underscore
 def test_issue4849():
     nlp = English()
 
-    # reset the Underscore object because test_underscore has a lambda function that can't be pickled
-    Underscore.doc_extensions = {}
-    Underscore.span_extensions = {}
-    Underscore.token_extensions = {}
-
     ruler = EntityRuler(
         nlp, patterns=[
             {"label": "PERSON", "pattern": 'joe biden', "id": 'joe-biden'},

spacy/tests/regression/test_issue4903.py

@@ -33,11 +33,6 @@ class CustomPipe:
 def test_issue4903():
     # ensures that this runs correctly and doesn't hang or crash on Windows / macOS
 
-    # reset the Underscore object because test_underscore has a lambda function that can't be pickled
-    Underscore.doc_extensions = {}
-    Underscore.span_extensions = {}
-    Underscore.token_extensions = {}
-
     nlp = English()
     custom_component = CustomPipe()
     nlp.add_pipe(nlp.create_pipe("sentencizer"))