Add pluggable server-side TextFilter service (social.md §4 US-4)

Deterministic default (ListBasedTextFilter) backed by a configurable JSON term list, with normalization (case-fold, leetspeak substitutions, stripped punctuation/spacing between letters) so trivial evasion doesn't slip through. Supports both reject (blocking) and redact (asterisk replacement) modes. TextFilter is a Protocol, leaving an explicit extension point for a future third-party moderation provider. Standalone module, not wired into any router/endpoint -- sibling chat tasks depend on this one. No DB/migration work. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Y2FESCaMXeJpC9gSctiEeD
2026-09-15 00:32:17 +00:00
parent f40940a491
commit 9334caf533
3 changed files with 518 additions and 0 deletions
@@ -0,0 +1,194 @@
+"""Unit tests for app.services.text_filter (docs/planning/social.md §4/§5 US-4).
+
+Pure, side-effect-free service -- no DB/Redis fixtures needed.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import pytest
+
+from app.services.text_filter import (
+    DEFAULT_TERMS_PATH,
+    FilterAction,
+    FilterMode,
+    FilterResult,
+    ListBasedTextFilter,
+    TextFilter,
+    default_text_filter,
+    load_terms,
+)
+
+
+@pytest.fixture
+def redact_filter() -> ListBasedTextFilter:
+    return ListBasedTextFilter(terms=["badword", "shoot"], mode=FilterMode.REDACT)
+
+
+@pytest.fixture
+def reject_filter() -> ListBasedTextFilter:
+    return ListBasedTextFilter(terms=["badword", "shoot"], mode=FilterMode.REJECT)
+
+
+class TestAllowPath:
+    def test_clean_text_is_allowed_unchanged(self, redact_filter: ListBasedTextFilter) -> None:
+        result = redact_filter.check("hello there, how are you today?")
+        assert result.action is FilterAction.ALLOW
+        assert result.filtered_text == "hello there, how are you today?"
+        assert result.matched_terms == ()
+
+    def test_message_with_no_matches(self, reject_filter: ListBasedTextFilter) -> None:
+        # Distinct from the clean-text case above: exercises the reject-mode
+        # filter's no-match path specifically (mode should not matter when
+        # nothing matches).
+        text = "let's meet at the plaza around noon"
+        result = reject_filter.check(text)
+        assert result.action is FilterAction.ALLOW
+        assert result.filtered_text == text
+        assert result.matched_terms == ()
+
+
+class TestRedactPath:
+    def test_exact_match_is_redacted(self, redact_filter: ListBasedTextFilter) -> None:
+        result = redact_filter.check("don't be a badword about it")
+        assert result.action is FilterAction.REDACT
+        assert result.matched_terms == ("badword",)
+        assert "badword" not in result.filtered_text
+        assert result.filtered_text == "don't be a ******* about it"
+
+    def test_multiple_distinct_matches(self, redact_filter: ListBasedTextFilter) -> None:
+        result = redact_filter.check("badword, don't shoot!")
+        assert result.action is FilterAction.REDACT
+        assert set(result.matched_terms) == {"badword", "shoot"}
+        assert "badword" not in result.filtered_text
+        assert "shoot" not in result.filtered_text
+        # Length is preserved (asterisks replace matched spans 1:1).
+        assert len(result.filtered_text) == len("badword, don't shoot!")
+
+    def test_redaction_char_is_configurable(self) -> None:
+        f = ListBasedTextFilter(terms=["badword"], mode=FilterMode.REDACT, redaction_char="#")
+        result = f.check("that's a badword thing to say")
+        assert result.filtered_text == "that's a ####### thing to say"
+
+
+class TestRejectPath:
+    def test_match_is_rejected_and_text_left_intact(self, reject_filter: ListBasedTextFilter) -> None:
+        text = "stop, badword!"
+        result = reject_filter.check(text)
+        assert result.action is FilterAction.REJECT
+        assert result.matched_terms == ("badword",)
+        # Reject mode does not mutate the text -- callers are expected to
+        # not store/deliver it at all, based on `action`.
+        assert result.filtered_text == text
+
+
+class TestNormalization:
+    def test_leetspeak_variant_is_caught(self, redact_filter: ListBasedTextFilter) -> None:
+        # "b4dw0rd" -> normalizes to "badword" via the leet map (4->a, 0->o).
+        result = redact_filter.check("you're being a b4dw0rd right now")
+        assert result.action is FilterAction.REDACT
+        assert result.matched_terms == ("badword",)
+
+    def test_separators_between_letters_are_stripped(self, redact_filter: ListBasedTextFilter) -> None:
+        result = redact_filter.check("that is b.a.d.w.o.r.d behavior")
+        assert result.action is FilterAction.REDACT
+        assert result.matched_terms == ("badword",)
+
+    def test_spacing_between_letters_is_stripped(self, redact_filter: ListBasedTextFilter) -> None:
+        result = redact_filter.check("b a d w o r d, seriously")
+        assert result.action is FilterAction.REDACT
+        assert result.matched_terms == ("badword",)
+
+    def test_case_folding(self, redact_filter: ListBasedTextFilter) -> None:
+        result = redact_filter.check("BADWORD in all caps")
+        assert result.action is FilterAction.REDACT
+        assert result.matched_terms == ("badword",)
+
+    def test_symbol_lookalikes(self) -> None:
+        # @ -> a, $ -> s: "@ss" should normalize to "ass".
+        f = ListBasedTextFilter(terms=["ass"], mode=FilterMode.REDACT)
+        result = f.check("what an @$$ move")
+        assert result.action is FilterAction.REDACT
+        assert result.matched_terms == ("ass",)
+
+
+class TestEmptyAndWhitespace:
+    def test_empty_string(self, redact_filter: ListBasedTextFilter) -> None:
+        result = redact_filter.check("")
+        assert result.action is FilterAction.ALLOW
+        assert result.filtered_text == ""
+        assert result.matched_terms == ()
+
+    def test_whitespace_only_string(self, redact_filter: ListBasedTextFilter) -> None:
+        text = "   \t\n  "
+        result = redact_filter.check(text)
+        assert result.action is FilterAction.ALLOW
+        assert result.filtered_text == text
+        assert result.matched_terms == ()
+
+    def test_punctuation_only_string_has_nothing_to_normalize(self, redact_filter: ListBasedTextFilter) -> None:
+        # No alnum characters survive normalization at all -- must not
+        # crash on an empty normalized string / empty index_map.
+        result = redact_filter.check("...!!!???")
+        assert result.action is FilterAction.ALLOW
+        assert result.matched_terms == ()
+
+
+class TestConfigurability:
+    def test_custom_term_list_is_used_instead_of_default(self) -> None:
+        f = ListBasedTextFilter(terms=["zzzcustomzzz"], mode=FilterMode.REDACT)
+        result = f.check("this contains zzzcustomzzz right here")
+        assert result.action is FilterAction.REDACT
+        assert result.matched_terms == ("zzzcustomzzz",)
+        # Default list terms should NOT match since we overrode the list.
+        result2 = f.check("this is fucking unrelated")
+        assert result2.action is FilterAction.ALLOW
+
+    def test_duplicate_and_blank_terms_are_deduped(self) -> None:
+        f = ListBasedTextFilter(terms=["badword", "badword", "  ", ""], mode=FilterMode.REDACT)
+        assert f._terms == ("badword",)
+
+    def test_default_terms_file_loads_and_is_nonempty(self) -> None:
+        terms = load_terms()
+        assert isinstance(terms, list)
+        assert len(terms) > 0
+        assert all(isinstance(t, str) and t for t in terms)
+
+    def test_default_terms_path_exists(self) -> None:
+        assert DEFAULT_TERMS_PATH.exists()
+        assert DEFAULT_TERMS_PATH.suffix == ".json"
+
+    def test_load_terms_from_explicit_path(self, tmp_path: Path) -> None:
+        custom = tmp_path / "custom_terms.json"
+        custom.write_text('{"terms": ["foo", "bar"]}', encoding="utf-8")
+        assert load_terms(custom) == ["foo", "bar"]
+
+    def test_default_text_filter_factory_matches_default_list(self) -> None:
+        f = default_text_filter(mode=FilterMode.REJECT)
+        result = f.check("that's fucking ridiculous")
+        assert result.action is FilterAction.REJECT
+        assert "fuck" in result.matched_terms
+
+
+class TestProtocol:
+    def test_list_based_filter_satisfies_text_filter_protocol(self, redact_filter: ListBasedTextFilter) -> None:
+        assert isinstance(redact_filter, TextFilter)
+
+    def test_check_returns_filter_result(self, redact_filter: ListBasedTextFilter) -> None:
+        result = redact_filter.check("hi")
+        assert isinstance(result, FilterResult)
+
+    def test_a_third_party_style_implementation_satisfies_the_protocol(self) -> None:
+        # Demonstrates the extension point: any object with a matching
+        # `check` method satisfies TextFilter without inheriting from
+        # anything -- this is how a future third-party provider would plug
+        # in without touching this module.
+        class AlwaysAllow:
+            def check(self, text: str) -> FilterResult:
+                return FilterResult(action=FilterAction.ALLOW, filtered_text=text, matched_terms=())
+
+        provider: TextFilter = AlwaysAllow()
+        assert isinstance(provider, TextFilter)
+        result = provider.check("anything at all, even fuck")
+        assert result.action is FilterAction.ALLOW