testing-and-quality-assurance（测试和质量保证）

Testing & Quality Assurance

Overview

Test infrastructure uses pytest. The tests/ directory has a mix of utility scripts and actual test_*.py files. tests/conftest.py and pytest.ini are the canonical test configuration.

Running Tests

# From project root with .venv active
.venv/bin/python -m pytest                    # run all tests
.venv/bin/python -m pytest -m unit            # unit tests only
.venv/bin/python -m pytest -m "not slow"      # skip slow ML tests
.venv/bin/python -m pytest tests/test_basic.py -v

pytest.ini Markers

[pytest]
testpaths = tests
addopts = -v --tb=short -q
markers =
    unit: Pure unit tests — no I/O or network
    integration: Tests that require a real database or network
    translation: Tests that exercise the Helsinki-NLP pipeline
    slow: Tests expected to take > 5 seconds

conftest.py Fixtures

tests/conftest.py provides:

Critical Patch Targets

The app uses module-level globals — patch them correctly:

# CORRECT — variable is dict_db, not app.db
from unittest.mock import patch, MagicMock

def test_search(client):
    mock = MagicMock()
    mock.search_entries.return_value = [{'chuukese_word': 'ran', 'english_translation': 'water'}]
    with patch('app.dict_db', mock):
        resp = client.get('/api/dictionary/search?q=ran')
    assert resp.status_code == 200

# CORRECT — collection attribute is dictionary_collection
mock_db.dictionary_collection.find.return_value = iter([...])

# CORRECT — method is bulk_insert_entries, not bulk_insert
mock_db.bulk_insert_entries.return_value = {'inserted': 5}

# CORRECT — translator is a module global, not a function
with patch('app.helsinki_translator') as mock_translator:
    mock_translator.translate.return_value = 'water'
    ...

API Test Pattern

import pytest

@pytest.mark.unit
def test_dictionary_search_empty_query(client, auth_headers):
    resp = client.get('/api/dictionary/search?q=')
    assert resp.status_code in (200, 400)

@pytest.mark.integration
def test_bible_coverage(client, auth_headers):
    resp = client.get('/api/database/bible-coverage')
    data = resp.get_json()
    assert 'books' in data

Translation Accuracy Testing

from sacrebleu.metrics import BLEU

bleu = BLEU()
result = bleu.corpus_score(hypotheses, [references])
assert result.score > 15.0, f"BLEU too low: {result.score}"

Test File Inventory

Files in tests/ prefixed test_ are runnable with pytest:

• test_basic.py — publication manager, JW.org lookup
• test_collections.py — database collection ops
• test_helsinki_trainer.py — HelsinkiNLP training helpers
• test_translation.py — translation pipeline
• test_scripture_parsing.py — scripture parsing logic
• test_word_families.py — word family grouping

Files without the test_ prefix are exploratory scripts and are not collected by pytest.

Source Files

• tests/conftest.py — shared fixtures
• pytest.ini — test runner configuration
• tests/test_basic.py — canonical example tests