asmello / grafq Goto Github PK

It would be very nice if the Schema class (and TypedField) could provide dynamic type hinting without us having to go through the static code generation. The dataclass module seems to do something like that.

At #4, I found out that there isn't really a way to do type checking statically while using dynamic code (sounds obvious, but dataclass does it, only by cheating!). This means that to improve the type validation experience, some form of static code must be used.

Two options available are:

1. Allow users to define their data classes, and provide thoughtful interfaces for interoperating with library code seamlessly. In other words, let users define something like dataclasses, and we only implement the logic behind their operation.
2. Fully generate the classes code from the GraphQL schema, by providing a script utility to users. This autogenerated codebase will have be be occasionally regenerated to pick up API changes, and cannot be customised (otherwise syncing will be a nightmare), but once in place, users can get the best out of their IDEs without writing any additional code.

There are obviously tradeoffs involved.

(1) has been used successfully for many ORM libraries (think SQLAlchemy), and is a very flexible approach that allows for interesting customisation opportunities. But this comes at the cost of delegating part of the work back to users - even though we know the types at runtime, they still have to, er, type them all up.

(2)'s approach been used successfully in some libraries like conjure-python. It requires breaking the glass a bit and instead of just importing this library, users will have to run a local script, and maintain their own snowflake code. Customisation is technically possible, but ill advised since they make automatically synchronising with the remote API non-trivial.

I suppose it doesn't have to be either-or, but I have to pick one to focus on first.