GeezerVison

Pull request type

• Code changes (bugfix, features)

Checklist

• Tests for the changes have been added (partially done)
• Docs have been reviewed and added / updated (future PR)
• Lint (black rocketpy/ tests/) has passed locally
• All tests (pytest tests -m slow --runslow) have passed locally
• CHANGELOG.md has been updated (if relevant)

Current behavior

Currently the _encoding module is just a stub and cannot still encode full simulations into a JSON like format.

New behavior

This PR expand this module bringing the following:

• Custom Encoding for more types;
• Customized behavior for some required classes with methods to_dict;
• Support for Function encoding.

This PR also introduces decoding functionalities. The way this works is:

• If the is a to_dict or from_dict method, they are used to perform the (de)encoding;
• Otherwise, a default encoding of all vars() and constructor based decoding is done.

Breaking change

• Yes
• No

Additional information

There is much to discuss here on the implementation and maintainability side of things. Main discussion points:

• Should we add a to_dict and (in the future) a from_dict method to some (or all) rocketpy classes?
  • My Opinion: I think adding the methods to everything might be the only way out. But there is some work and maintainability considerations.
• The current status of the encoding can be fully deserialized?
  • My Opinion: All the information is there, but the typing should be improved to know which objects to decode (deserialize) into. For that reason, I kept the _encoders private.
• Are there any other modules to make our work easier?
  • From my research, I have tested the following:
    • simplejson: brings some more types and speed into encoding, but does not help much with custom types;
    • jsonpickle: really great for the typing handler (already writes by default each object type) and is able to serialize. The problem is that is does not handle functions/lambdas and I did not find a good way to implement custom handlers for our use case;
    • pickle / dill: the most plug and play solution, however the output is not human readable and not generally compatible between different rocketpy versions.

Of course, this is what I understood upon researching and testing, feel free to make any comments or suggest other modules.

Remarks and Future Considerations

Here are some of the steps that could be improved:

• Verify why the test for hybrid motor flight requires higher tolerances;
  • Corrected, no higher tolerances needed anymore.
• Test if all of the atmospheric models of Environments can be decoded without requiring re-read of files or API;
• Evaluate whether we should encode outputs or only the minimum required information to decode the class from its init;
  • Option for including outputs or not was added.
• (Optional) Expand implementation to AirBrakes + Controllers and new classes.

Some important remarks:

• The simple getting started simulation is essentially identical comparing the store & load, the only difference is the random lag and noise from Parachute class. I could not found a simple solution to reset those and always get the same results.