* Add collections & indexes overrides for Fauna DB
* Fix the name of the verification token index
Co-authored-by: Florian Michaut <florian@coding-days.com>
* Centralises configuration for Puppeteer used in tests to make it easier to maintain.
* Adds support for running tests on ARM, so we can use Raspberry Pi test runners off the cloud to get around block lists.
* Includes improved stealth mode to avoid detection which breaks integration tests.
Full end-to-end integration tests for Twitter (OAuth 1) and GitHub (OAuth 2) using Puppeteer and Mocha.
This replaces Cypress tests due to issues with Cypress not being able to run tests against external URLs, which we need for our integration tests.
The integration test runner is hosted outside of GitHub Actions (it cannot be hosted by GitHub or on AWS due to IP access controls placed on sign in by providers like Twitter and GitHub) and so the integration tests may not pass if the test runner is offline. If this happens, tests can be re-run later when the test runner is available.
See Pull Request #641 for details.
Adds commands to start/rebuild/stop a Docker image of a sample Next.js app that loads the latest build of NextAuth.js from the current directory.
* `npm run test:app:start`
* `npm run test:app:rebuild`
* `npm run test:app:stop`
It is intended for further development for automated testing.
### About the build process
* The Dockerfile uses a multi-stage build process to optimise build performance, but the nature of the process is slow.
* Build times vary depending on computer speed and internet connection.
* Inital build times are slow (it may take 10 minutes or more).
* Subsequent builds on the same computer should be faster (1 minute or less).
* To ensure the package.json is valid, modules required in the next-auth package.json file are re-downloaded* on every build.
* A Docker compose file is used to allow us to extend the test app to run it again multiple databases.
Subsequent updates may look to improve performance, but it's important checks like checking package.json is valid and running the build in isolation are performed.
An issue with the defaults for MySQL used by TypeORM Adapter has been highlighted during testing parity with the Prisma Adapter.
This change ensures *all* TIMESTAMP columns use TIMESTAMP(6) to store six digits of precision after the number of seconds.
While this is level of precision is not required everywhere it ensures all timestamps in the default models use the same configuration (instead of a mix of values) and is consisitant with the level of precision on timestamps used Postgres.
