CUDless testing

Recipe for efficient testing with pytest for flask-sqlalchemy based applications.

FAQ

• What is this?

  I had two visions for efficient testing:

  • I want to write test suites which can utilize a database but never have to actually write (i.e. commit) rows to it.
  • I don't want to setup/initialize the database with its default table rows (e.g. a table of all country names or a table of available color names) for every single test run. Instead, I want to initialize the database once and then run all my tests on top of that.

  This recipe is an effort in accomplishing these two goals.

• Can you briefly explain how this recipe works?

  • All the testing happens within a main transaction which is rolled back at the end of testing.
  • I use nested subtransactions to still be able to commit and rollback freely in my application code.
  • The database initialization is performed in a lower nesting level than the one in which the individual tests are run in. Therefore, after each individual test, all changes performed by that test can be just rolled back without the initialization tasks being rolled back as well (hence the database initialization is isolated from the tests).

• How do I use this for my own project?

  The most important file is tests/conftest.py. This is where all the heavy lifting is done. I suggest that you copy this file to your own tests directory and then read through it and modify it as necessary.

  tests/test_example.py is an example of what your tests could look like.

  cudless_testing/__init__.py is an example of what your application code could look like.

• Which database dialects are supported?

  I have tested it with SQLite only but there is nothing dialect specific in the code which would prevent you from using a different dialect. Keep in mind though that support for SAVEPOINTs is required.

• How is this recipe different to SQLAlchemy's recipe in the docs?

  While SQLAlchemy's recipe works great, I think it's a bit inconvenient to bind a Session to a Connection which is already in a transactional state, when using the flask-sqlalchemy library.

• Why CUDless?

  Since this recipe is utilizing a database without creating, updating or deleting any rows in a committed transaction, we use the acronym CRUD but without the R for read. And because we never actually commit any CUD, our approach is "CUDless".