Updates for VS Code

[JamesHutchison]

Background

The library has sat for a while and consequentially no longer was working "out of the box" with VS Code. After doing some investigation, determined the following:

• VS Code's latest implementation doesn't pass the items under test to the application anymore. Rather, it calls pytest.main directly
• The test runner now blocks (and spins) if it detects a test process is still running
• VS Code communicates which tests to run via an environment variable
• VS Code injects their own plugin as the first argument passed into pytest. The path is modified to allow running with this plugin

This PR updates everything VS Code. In addition to fixing the ability to run tests, it also updates the dev container configuration.

Note that there is no solution to the test process issue. I noticed that when a test that starts the daemon runs, there's actually two processes that persist, the parent client and the child daemon. However, I noticed that killing the client didn't satisfy VS Code, and it was only satisfied if the daemon was killed as well. Having the tests hang is a bad experience, so I added a parameter for automatically starting the daemon, and now the client will simply error if the server is not running. I plan on raising this issue with the VS Code Python maintainers to see if there's a way we can get these two to cooperate. You can bypass the new logic in VS Code by adding "python.experiments.optOutFrom": [ "pythonTestAdapter" ] to your config.

While double checking whether the hot reloading functionality still worked as expected, I noticed the breakpoint line numbers were no longer lining up like they used to. Jurigged had a bug where functions would have the line numbers drift versus the real ones, and a (large) monkey patch was added to correct the result of the logic. Using the latest version of the library, it appears the monkey patch is now causing the drift, possibly because the issue was fixed since I last checked. I'll investigate this and alter or remove the monkey patch in a different PR. https://github.com/JamesHutchison/pytest-hot-reloading/issues/51 was created

^ Upon investigating it looks like I may have not paid close enough attention and what I thought were breakpoints not hitting were actually it failing to get past MegaPatch.it(...) in the test.

Dev Container Changes

• Added mypy extension (not the official one, which sucks)
• Switched formatting provider to ruff
• The pytest daemon is now automatically started in the background on dev container start

Client Changes

• The client now uses the pytest config's invocation_params.dir instead of cwd() for giving the current working directory to the server. I'm not sure this changes anything but it seems prudent to do
• The client now uses the pytest config's invocation_params.args instead of logic that extracted the params from the command line parameters. This was needed because VS Code no longer was passing the test via the command line and actually I'm really glad this change was made because it cut out a bunch of complexity and lines of code
• Client now sends the environment and current path to the server
• The amount of time the client waits for the server to start was increased. Previously it would give up after about a second, now it should give up after about 10.

Server Changes

• Now takes in environment in JSON form. JSON is used because it was getting serialization errors in some cases when passing a dict
• Now takes the path from the client

Plugin Changes

• If pytest --help is used, then it will no longer do plugin logic. This was needed because the exception raised by the server not running was a problem.
• Added --daemon-start-if-needed parameter with corresponding PYTEST_DAEMON_START_IF_NEEDED environment variable