NRCL-103 Require patched NLTK in the optional sentiment extra

pyproject.toml

@@ -83,6 +83,7 @@ fix = [
 ]
 sentiment = [
     "textblob>=0.17.1",
+    "nltk>=3.10.3",
     "vaderSentiment>=3.3.2",
     "transformers>=4.21.0",
     "torch>=1.12.0",

tests/test_sentiment_textblob.py

@@ -0,0 +1,33 @@
+"""Optional TextBlob smoke coverage without downloaded corpora or network access."""
+
+import socket
+
+import pytest
+
+
+@pytest.mark.parametrize(
+    ("text", "direction"),
+    [("I love this excellent result.", 1), ("I hate this terrible result.", -1)],
+)
+def test_textblob_sentiment_without_network_or_corpora(monkeypatch, text, direction):
+    def forbid_network(*args, **kwargs):
+        raise AssertionError("Sentiment analysis must not access the network or download corpora")
+
+    monkeypatch.setattr(socket.socket, "connect", forbid_network)
+    monkeypatch.setattr(socket.socket, "connect_ex", forbid_network)
+    monkeypatch.setattr(socket, "create_connection", forbid_network)
+    monkeypatch.setattr(socket, "getaddrinfo", forbid_network)
+
+    nltk = pytest.importorskip("nltk")
+    monkeypatch.setattr(nltk.data, "path", [])
+    monkeypatch.setattr(nltk, "download", forbid_network)
+    pytest.importorskip("textblob")
+
+    from neural.analysis.sentiment import SentimentAnalyzer, SentimentEngine
+
+    score = SentimentAnalyzer(engine=SentimentEngine.TEXTBLOB).analyze_text(text)
+
+    assert "textblob" in score.metadata["engines_used"]
+    assert score.engine_used is SentimentEngine.TEXTBLOB
+    assert 0 < direction * score.overall_score <= 1
+    assert 0 <= score.subjectivity <= 1