Library for guest-host data exchange

[dogtopus]

Write demos for each one of the candidates listed below and evaluate them based on the requirements

Requirements

Preferably we want the following:

• Basic field serialization/deserialization (of course)
• Enum handling. Integrating Python Enum/IntEnum would be a plus.
• Easy integration into existing typing and IDE setup.
• Handles alignment well.
• Ways to easily wrap guest pointer access. (Optional but nice to have. Though I'm not sure on this since then everything needs to depend on Unicorn. Maybe a function and a Protocol that allows reusable data reading/writing would be a solution?)
• Reasonably good performance.
• Play well with OO paradigm may not be a requirement, but comes up as natural when there are multiple places that require changing different fields in a coherent way, and, specifically for Python, one-off interpretation of some fields (OO, PO, functional are all able to reuse code pretty well so reusability won't be a concern, but OO naturally bundles code with object so they can be accessed more conveniently on Python).

Candidates

Currently there are 3 candidates:

• Built-in struct module + dataclass
  • Currently in use.
  • Fast (at least for struct itself).
  • Works reasonably well but for larger structs with helper class for processing one needs to manage 2 separate sets of schema (both in struct format string which is hard to maintain, and dataclass).
• Construct
  • Very mature declarative parser.
  • Fast.
  • May have similar issue as struct as we may need to maintain 2 sets of schema if typing and data processing is required.
• Caterpillar
  • Very new and potentially unstable.
  • Slower than construct but should not be a big concern from dev's test results.
  • Cleanest interface so far. Should play well with what we have now.

[dogtopus]

An alternative option to this would be write our own dataclass-like declarative parser for data exchange. This way we can seamlessly support unicorn and do things like nested struct parsing and lazy access. Both Caterpillar and dataclasses offer some ideas on how to achieve this.