This change introduces PostgreSQL compatibility, by implementing the
following changes:
* Use `JSONB` instead of `JSON` for PostgreSQL databases. This is
achieved by creating a custom `CustomJSON` type that uses `JSONB` on
PostgreSQL and `JSON` on other databases, leveraging the variant
mechanism of SQLAlchemy.
* Add `is_postgresql` function to check if the current database is
PostgreSQL. Commonly used for migrations, but can be used to determine
how to build queries based on the database engine.
* Add `json_array_contains_value` function to check if a JSON array
includes a specific value. This function is needed as it's
engine-specific.
Support for PostgreSQL is on a best-effort basis, and it relies on the
community for continued support and testing. The project's main database
is MariaDB.
Closes#667.
The key `romm:twitch_token_expires_at` is not needed, as we already
set an expiration time for the token key itself. If the token is still
present in Redis, it is valid. If not, it means we either never fetched
it, or it was too close to expiration.
With this change, fetching the Twitch token makes a single Redis call
instead of two.
The `get_roms` method is used during scanning and to generate feeds.
Sorting by filename is not perfect (e.g. prefixes like "The" or "A"),
but should be good enough for users to better visualize how the scanning
process is going, and how close it's to finish.
The scanning process could try to fetch thousands of roms from the
database, one by one, which is very inefficient, especially when only a
few new roms are added to the library, as the overhead of database calls
is very high compared to metadata API calls.
This change introduces a new method in the roms handler to retrieve roms
by their file names in bulk, which is used during the scan process to
fetch a batch of roms at once, instead of one by one.
It also avoids the `@with_details` decorator when fetching roms during
the scanning process, as those details are not needed, and they were
adding unnecessary joins and data decoding.
