Skip to content

testing.md

Latest commit

 

History

History
140 lines (106 loc) · 4.33 KB

testing.md

File metadata and controls

140 lines (106 loc) · 4.33 KB
Raw

Testing

Run the Suite

cd backend
PYTHONPATH=. /home/container/python3.11/python/bin/python3.11 -m pytest tests/ -v

Expected output:

tests/test_fixes.py  18 passed in ~2s

Config lives in backend/pytest.ini:

[pytest]
asyncio_mode = auto   # async fixtures/tests need no decorators
testpaths = tests

What's Covered (18 tests)

TestAuthRotation — refresh token rotation

Test Proves
test_refresh_token_rotation_rejects_replay First refresh rotates; replaying the OLD token → 401 "invalidated"; new token works; exactly 1 session row remains

TestAuthRotation — download tokens

Test Proves
test_download_token_binds_file_id Token minted for file 42 authorizes file 42
test_wrong_file_id_rejected Same token on file 43 → 403 (IDOR blocked)
test_malformed_sub_returns_none JWT with sub="notanint" → None, no crash

TestParseRangeHeader — HTTP Range parsing (8 cases)

Test Input → Result
no header None → full content (200)
normal bytes=0-99 → (0, 99)
open-ended bytes=100- → (100, EOF)
suffix bytes=-500 → last 500 bytes
clamped end end > size → clamped to size-1
multipart rejected bytes=0-1,5-6 → single coalesced range, not multi-part
zero-byte + GET size=0, no header → full 200
zero-byte + Range size=0 with header → handled without 416 crash

TestPatchListenerQueue — bot listener handling

Test Proves
test_fifo_delivery Two replies arrive out of order → listener gets them in arrival order, both delivered
test_command_not_swallowed /start from a user with a pending listener is routed as a command, not eaten by the listener

TestDiskCacheUniqueTemp

Test Proves
test_concurrent_put_no_tear 8 concurrent put() calls → all 8 files intact, no temp-file clobbering

TestGDriveRawWrite

Test Proves
test_raw_write_creates_file _raw_write creates the file at the right path

TestGrabberCollectReplies

Test Proves
test_direct_reply_preferred Direct reply to request beats other candidates

How Isolation Works

Each test gets a fresh SQLite DB in its own tmpdir:

temp_db fixture:
  1. tempfile.mkdtemp()                    ← unique per test
  2. Set DATABASE_URL to sqlite+aiosqlite:///<tmpdir>/t.db
  3. importlib.reload(app.config / database / models)
  4. create_all()                          ← fresh schema
  5. yield AsyncSession
  6. dispose engine + shutil.rmtree(tmpdir)

This is why tests can run in any order and never see each other's rows (the old shared-DB approach hit UNIQUE constraint failed: users.telegram_id).

Env vars needed by app imports are set in tests/conftest.py before any app.* import — dummy Telegram tokens etc. No real credentials are ever touched.

Adding a Test

@pytest.mark.asyncio
async def test_my_feature(self, temp_db):
    from app.models import User
    user = User(telegram_id=111, username="u", auth_version=0)
    temp_db.add(user)
    await temp_db.commit()

    # call the function under test...

Rules of thumb:

  1. Need a DB? Take the temp_db fixture (it is the session).
  2. Need a user? Take sample_user instead.
  3. Pure function? No fixture needed (parse_range_header tests).
  4. Telegram-dependent code? Monkeypatch the client object — never hit the network.
  5. Keep one behavior assertion-cluster per test; name it after the behavior.

What's NOT Covered (gaps)

  • Live streaming pipeline (needs real Telegram) — covered by manual diag checks instead: 
    curl -H "Authorization: Bearer $DEBUG_PASSWORD" \
  "http://localhost:7680/api/diag/stream?msg=197&chat=-1003950847652" \
  -H "Range: bytes=0-1023" -o /dev/null -w "%{http_code} %{size_download}\n"
# expect: 206 1024
  • Frontend (no JS test runner configured yet)
  • Subtitle providers (external network)

CI Notes

No CI configured yet. Minimum gate before pushing:

cd backend && PYTHONPATH=. python3.11 -m pytest tests/ -q \
  && python3.11 -m compileall -q app
cd frontend && npx tsc --noEmit && npx eslint . --ext ts,tsx --max-warnings 0