The SQL statement maps have traditionally been hard-coded dictionaries

of strings/lists of strings, accessible when importing the module. Over

time, these became a little more complex, handling variations in

statements based on the database version or Django version, but they

remained static dictionaries.
The problem with them being static is that it wasn't feasible to run the

unit tests againt different versions of MySQL, MariaDB, or Postgres

databases. The SQL was generated dynamically based on the capabilities

of a named database connection, like mysql_innodb, and that meant that

a new MySQL-based database couldn't be registered for testing an

alternate version.
To enable these more version-specific tests, all mappings are now

generated dynamically when needed. Instead of static dictionaries,

they're functions that take a database connection and return a mapping

dictionary. These functions can generate SQL suitable for that

connection without having to hard-code anything about the connection.
The resulting SQL maps are cached during the test run, so that they only

need to be generated once.
With this change, a test_db_settings.py is now free to define as many

databases as needed, and any of those can be passed to the test runner.

The template or this file has been updated with examples.

Unit tests pass on all versions of Django for all supported databases,

plus for additional database entries (tested with MariaDB, MySQL 5.6,

MySQL 5.7, and MySQL 8).