Spaces:
Running
Running
Download tests/test_source_resolution.py from anish0634/ARIA-Agentic-chatbot: direct link, hf CLI and curl.
- Browser
- Download file 9.89 kB
-
https://huggingface.co/spaces/anish0634/ARIA-Agentic-chatbot/resolve/main/tests/test_source_resolution.py
- Command line
-
hf download hf://spaces/anish0634/ARIA-Agentic-chatbot/tests/test_source_resolution.py
-
curl -L -o test_source_resolution.py https://huggingface.co/spaces/anish0634/ARIA-Agentic-chatbot/resolve/main/tests/test_source_resolution.py
9.89 kB
| """ | |
| Tests for the source-resolution / broad-query feature in app/services/rag.py. | |
| Covers the class of question pure semantic similarity can't answer: "what is | |
| in this document", "what's in the excel", "what is in this chat/inbox/mail", | |
| and synthesis-style asks ("key points", "risks") that aren't phrased verbatim | |
| in the source text. | |
| No LLM calls. Vector-store-touching helpers (list_session_sources, | |
| _fetch_broad_context) use a MagicMock store, same pattern as | |
| tests/test_clarification.py. | |
| """ | |
| import pytest | |
| from unittest.mock import MagicMock | |
| from langchain_core.documents import Document | |
| from app.services.rag import ( | |
| is_broad_query, | |
| _resolve_target_sources, | |
| list_session_sources, | |
| _fetch_broad_context, | |
| _fetch_all_docs_and_metas, | |
| ) | |
| # ββ is_broad_query ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| class TestIsBroadQuery: | |
| def test_detects_broad_intent(self, q): | |
| assert is_broad_query(q) is True | |
| def test_does_not_flag_specific_questions(self, q): | |
| assert is_broad_query(q) is False | |
| # ββ _resolve_target_sources βββββββββββββββββββββββββββββββββββββββββββββββββββ | |
| class TestResolveTargetSources: | |
| def setup_method(self): | |
| self.sources = [ | |
| {"source": "Food-Safety-Policy-01.04.19.pdf", "connector": None}, | |
| {"source": "budget_report.xlsx", "connector": None}, | |
| {"source": "Google Chat - team-standup.txt", "connector": "google_chat"}, | |
| {"source": "Project kickoff (abcd1234).txt", "connector": "gmail"}, | |
| ] | |
| def test_no_sources_returns_none(self): | |
| assert _resolve_target_sources("what is in this", []) is None | |
| def test_filename_substring_match(self): | |
| result = _resolve_target_sources("what is in the food safety policy", self.sources) | |
| assert result == ["Food-Safety-Policy-01.04.19.pdf"] | |
| def test_filename_match_ignores_separator_punctuation(self): | |
| """Filenames use underscores ('budget_report.xlsx'); users type spaces.""" | |
| result = _resolve_target_sources("summarize the budget report", self.sources) | |
| assert result == ["budget_report.xlsx"] | |
| def test_extension_alias_excel(self): | |
| result = _resolve_target_sources("what is in this excel", self.sources) | |
| assert result == ["budget_report.xlsx"] | |
| def test_connector_alias_chat(self): | |
| result = _resolve_target_sources("what is in this chat", self.sources) | |
| assert result == ["Google Chat - team-standup.txt"] | |
| def test_connector_alias_space(self): | |
| result = _resolve_target_sources("what is in this space", self.sources) | |
| assert result == ["Google Chat - team-standup.txt"] | |
| def test_connector_alias_inbox_and_mail(self): | |
| for q in ["what is in my inbox", "what is in this mail", "what's in this email"]: | |
| assert _resolve_target_sources(q, self.sources) == ["Project kickoff (abcd1234).txt"] | |
| def test_demonstrative_with_multiple_sources_is_ambiguous(self): | |
| # "this document" with 4 distinct sources in the session can't be | |
| # safely resolved to one - must return None (normal/ambiguity-aware | |
| # retrieval should take over), not silently pick one. | |
| assert _resolve_target_sources("what is in this document", self.sources) is None | |
| def test_demonstrative_with_single_source_resolves(self): | |
| single = [{"source": "xyz_report.pdf", "connector": None}] | |
| assert _resolve_target_sources("what is in this document", single) == ["xyz_report.pdf"] | |
| assert _resolve_target_sources("what is in this file", single) == ["xyz_report.pdf"] | |
| def test_distinctive_token_fallback_single_source(self): | |
| """'xyz' in the question matches 'xyz_report.pdf' via its stem token, | |
| even though the full stem ('xyz report') isn't a substring of the | |
| question - this is the exact "filename xyz" case from the bug report.""" | |
| single = [{"source": "xyz_report.pdf", "connector": None}] | |
| assert _resolve_target_sources("what is in this xyz", single) == ["xyz_report.pdf"] | |
| def test_generic_stem_words_dont_false_positive(self): | |
| multi = [ | |
| {"source": "xyz_report.pdf", "connector": None}, | |
| {"source": "abc_summary.docx", "connector": None}, | |
| ] | |
| # "report" alone is a generic word shared by many filenames - must | |
| # not resolve to a specific source just because it appears in both. | |
| assert _resolve_target_sources("what is in this report", multi) is None | |
| def test_unrelated_question_returns_none(self): | |
| assert _resolve_target_sources("what is the weather today", self.sources) is None | |
| def test_multiple_named_files_scopes_to_all(self): | |
| result = _resolve_target_sources( | |
| "compare the food safety policy and the budget report", self.sources | |
| ) | |
| assert set(result) == {"Food-Safety-Policy-01.04.19.pdf", "budget_report.xlsx"} | |
| # ββ list_session_sources / _fetch_broad_context (vector-store-touching) ββββββ | |
| class TestFetchAllDocsAndMetas: | |
| def test_chroma_backend(self): | |
| vs = MagicMock() | |
| vs._collection = MagicMock() | |
| vs._collection.get.return_value = { | |
| "documents": ["a", "b"], | |
| "metadatas": [{"source": "x.pdf"}, {"source": "y.pdf"}], | |
| } | |
| docs, metas = _fetch_all_docs_and_metas(vs) | |
| assert docs == ["a", "b"] | |
| assert metas == [{"source": "x.pdf"}, {"source": "y.pdf"}] | |
| def test_pgvector_backend_no_collection_name_returns_empty(self): | |
| vs = MagicMock(spec=[]) # no _collection, no collection_name attr | |
| docs, metas = _fetch_all_docs_and_metas(vs) | |
| assert docs == [] | |
| assert metas == [] | |
| class TestListSessionSources: | |
| def _make_vs(self, rows: list[dict]): | |
| vs = MagicMock() | |
| vs._collection = MagicMock() | |
| vs._collection.get.return_value = { | |
| "documents": [f"chunk {i}" for i in range(len(rows))], | |
| "metadatas": rows, | |
| } | |
| return vs | |
| def test_deduplicates_sources(self): | |
| vs = self._make_vs([ | |
| {"source": "a.pdf", "chunk_index": 0}, | |
| {"source": "a.pdf", "chunk_index": 1}, | |
| {"source": "b.pdf", "chunk_index": 0}, | |
| ]) | |
| result = list_session_sources(vs) | |
| assert sorted(s["source"] for s in result) == ["a.pdf", "b.pdf"] | |
| def test_carries_connector_metadata(self): | |
| vs = self._make_vs([ | |
| {"source": "Google Chat - x.txt", "connector": "google_chat"}, | |
| ]) | |
| result = list_session_sources(vs) | |
| assert result[0]["connector"] == "google_chat" | |
| def test_empty_store_returns_empty_list(self): | |
| vs = self._make_vs([]) | |
| assert list_session_sources(vs) == [] | |
| def test_chunks_with_no_source_are_skipped(self): | |
| vs = self._make_vs([{"chunk_index": 0}]) | |
| assert list_session_sources(vs) == [] | |
| class TestFetchBroadContext: | |
| def _make_vs(self, rows: list[tuple[str, dict]]): | |
| vs = MagicMock() | |
| vs._collection = MagicMock() | |
| vs._collection.get.return_value = { | |
| "documents": [r[0] for r in rows], | |
| "metadatas": [r[1] for r in rows], | |
| } | |
| return vs | |
| def test_scopes_to_target_source_in_order(self): | |
| rows = [ | |
| ("chunk 2", {"source": "a.pdf", "chunk_index": 2}), | |
| ("chunk 0", {"source": "a.pdf", "chunk_index": 0}), | |
| ("other", {"source": "b.pdf", "chunk_index": 0}), | |
| ("chunk 1", {"source": "a.pdf", "chunk_index": 1}), | |
| ] | |
| vs = self._make_vs(rows) | |
| docs = _fetch_broad_context(vs, "sess", ["a.pdf"], limit=10) | |
| assert [d.page_content for d in docs] == ["chunk 0", "chunk 1", "chunk 2"] | |
| def test_respects_limit(self): | |
| rows = [(f"c{i}", {"source": "a.pdf", "chunk_index": i}) for i in range(10)] | |
| vs = self._make_vs(rows) | |
| docs = _fetch_broad_context(vs, "sess", ["a.pdf"], limit=3) | |
| assert len(docs) == 3 | |
| def test_no_target_spreads_across_sources(self): | |
| rows = ( | |
| [(f"a{i}", {"source": "a.pdf", "chunk_index": i}) for i in range(5)] + | |
| [(f"b{i}", {"source": "b.pdf", "chunk_index": i}) for i in range(5)] | |
| ) | |
| vs = self._make_vs(rows) | |
| docs = _fetch_broad_context(vs, "sess", None, limit=4) | |
| sources_seen = {d.metadata["source"] for d in docs} | |
| # Must not be starved to just one source when nothing was named. | |
| assert sources_seen == {"a.pdf", "b.pdf"} | |
| def test_empty_store_returns_empty_list(self): | |
| vs = self._make_vs([]) | |
| assert _fetch_broad_context(vs, "sess", None) == [] | |
| def test_unknown_target_source_returns_empty(self): | |
| rows = [("c0", {"source": "a.pdf", "chunk_index": 0})] | |
| vs = self._make_vs(rows) | |
| assert _fetch_broad_context(vs, "sess", ["nonexistent.pdf"]) == [] | |