curo1305
21ec9cb4c3
- Add auth_user, admin_user, mock_minio_presigned, mock_minio_stat fixtures to conftest.py - Create test_quota.py with 4 xfail stubs (STORE-03, STORE-05, STORE-06, SC2 race) - Append test_migration_0003 to test_alembic.py (full pre-seed + post-migration assertions) - Append 3 classifier xfail stubs (DOC-03, DOC-05, D-15) - Append 6 document xfail stubs (D-05, STORE-04, SEC-04, D-16) - Append 4 topic xfail stubs (DOC-04, D-09, D-17) - Append test_settings_endpoint_removed stub (D-12) - All 19 new test IDs collect cleanly with xfail(strict=False)
149 lines
4.9 KiB
Python
149 lines
4.9 KiB
Python
"""
|
|
Unit tests for AI provider JSON parsing robustness and classifier orchestration.
|
|
Uses a mock provider — no real AI calls made.
|
|
"""
|
|
import json
|
|
import pytest
|
|
from ai.openai_provider import _parse_classification, _parse_suggestions, _strip_code_fences
|
|
from ai.base import ClassificationResult
|
|
|
|
|
|
def test_parse_clean_json():
|
|
raw = '{"assigned_topics": ["finance", "invoices"], "new_topic_suggestions": []}'
|
|
result = _parse_classification(raw)
|
|
assert result.topics == ["finance", "invoices"]
|
|
assert result.suggested_new_topics == []
|
|
|
|
|
|
def test_parse_with_code_fence():
|
|
raw = '```json\n{"assigned_topics": ["legal"], "new_topic_suggestions": ["contracts"]}\n```'
|
|
result = _parse_classification(raw)
|
|
assert result.topics == ["legal"]
|
|
assert result.suggested_new_topics == ["contracts"]
|
|
|
|
|
|
def test_parse_with_preamble():
|
|
raw = 'Here is the classification:\n{"assigned_topics": ["hr"], "new_topic_suggestions": []}\nDone.'
|
|
result = _parse_classification(raw)
|
|
assert result.topics == ["hr"]
|
|
|
|
|
|
def test_parse_malformed_returns_empty():
|
|
raw = "I cannot classify this document."
|
|
result = _parse_classification(raw)
|
|
assert result.topics == []
|
|
assert result.suggested_new_topics == []
|
|
|
|
|
|
def test_strip_code_fences():
|
|
raw = "```json\n{}\n```"
|
|
assert _strip_code_fences(raw) == "{}"
|
|
|
|
|
|
def test_parse_suggestions_clean():
|
|
raw = '{"suggested_topics": ["Human Resources", "Onboarding"]}'
|
|
result = _parse_suggestions(raw)
|
|
assert "Human Resources" in result
|
|
assert "Onboarding" in result
|
|
|
|
|
|
def test_parse_suggestions_with_fence():
|
|
raw = "```\n{\"suggested_topics\": [\"Finance\"]}\n```"
|
|
result = _parse_suggestions(raw)
|
|
assert result == ["Finance"]
|
|
|
|
|
|
def test_parse_suggestions_malformed():
|
|
raw = "No suggestions available."
|
|
result = _parse_suggestions(raw)
|
|
assert result == []
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_classifier_with_mock_provider(isolated_data_dir):
|
|
"""Test classifier orchestration with a mock provider."""
|
|
from unittest.mock import AsyncMock, patch
|
|
from ai.base import ClassificationResult
|
|
import services.storage as st
|
|
|
|
# Create a document
|
|
doc_id = "test-doc-1"
|
|
st.save_metadata({
|
|
"id": doc_id,
|
|
"original_name": "test.txt",
|
|
"filename": "test-doc-1.txt",
|
|
"mime_type": "text/plain",
|
|
"size_bytes": 50,
|
|
"extracted_text": "Invoice for services rendered in March 2026.",
|
|
"topics": [],
|
|
"created_at": "2026-01-01T00:00:00Z",
|
|
"classified_at": None,
|
|
})
|
|
|
|
# Create some topics
|
|
st.create_topic("Finance")
|
|
st.create_topic("Legal")
|
|
|
|
mock_result = ClassificationResult(
|
|
topics=["Finance"],
|
|
suggested_new_topics=["Invoices"],
|
|
reasoning="Document is about financial invoicing.",
|
|
)
|
|
|
|
with patch("services.classifier.get_provider") as mock_get_provider:
|
|
mock_provider = AsyncMock()
|
|
mock_provider.classify = AsyncMock(return_value=mock_result)
|
|
mock_get_provider.return_value = mock_provider
|
|
|
|
from services.classifier import classify_document
|
|
topics = await classify_document(doc_id)
|
|
|
|
assert "Finance" in topics
|
|
assert "Invoices" in topics
|
|
|
|
# Verify new topic was auto-created
|
|
all_topics = st.load_topics()
|
|
assert any(t["name"] == "Invoices" for t in all_topics)
|
|
|
|
# Verify document was updated
|
|
meta = st.get_metadata(doc_id)
|
|
assert "Finance" in meta["topics"]
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Wave 0 xfail stubs for per-user AI provider resolution — Plan 03-04
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
@pytest.mark.xfail(strict=False, reason="implemented in plan 03-04")
|
|
async def test_per_user_provider(db_session):
|
|
"""When user.ai_provider='openai' and user.ai_model='gpt-4o', the classifier
|
|
resolves _settings['active_provider'] == 'openai'.
|
|
|
|
DOC-03: AI provider/model comes from the user's DB record, not from global
|
|
config or the retired load_settings() flat file (CONTEXT.md D-14).
|
|
"""
|
|
assert True # scaffold
|
|
|
|
|
|
@pytest.mark.xfail(strict=False, reason="implemented in plan 03-04")
|
|
async def test_celery_task_uses_user_provider(db_session):
|
|
"""Calling _run(document_id) for a Document owned by user.ai_provider='anthropic'
|
|
calls classifier with ai_provider='anthropic'.
|
|
|
|
DOC-05: the Celery extract_and_classify task resolves per-user AI config via
|
|
a second DB lookup (doc.user_id → user.ai_provider/ai_model) and passes it
|
|
to the classifier (CONTEXT.md D-14).
|
|
"""
|
|
assert True # scaffold
|
|
|
|
|
|
@pytest.mark.xfail(strict=False, reason="implemented in plan 03-04")
|
|
async def test_default_provider_fallback(db_session):
|
|
"""When user.ai_provider is None, the classifier receives config.settings.default_ai_provider.
|
|
|
|
D-15: fallback chain is user.ai_provider → DEFAULT_AI_PROVIDER env var →
|
|
code default 'ollama' (CONTEXT.md D-15).
|
|
"""
|
|
assert True # scaffold
|