feature request/concept: statically-typed+reexecutable alternative to trustfall_stubgen

[zombiepigdragon]

First of all, I should preface this with the fact that I haven't had significant experience with trustfall, so I can't state with confidence that my conceptual usage pattern is correct or that it would work well. However, at my early point in the learning curve, I find myself wishing that the crate worked in the way I describe here.

My primary concern that leads to this feature request is that trustfall_stubgen generates some stringly-typed code, and is generally only run once to bootstrap the adapter. This works well if the schema is written prior to writing the adapter, but it is less well-suited to cases where the schema may change (the schema changes have to be manually transferred to the adapter without the tool's assistance, with check_adapter_invariants detecting the changes when the test suite is run rather than on the next build). However, I believe a better solution would be to generate a "adapter" type (in a way similar to trustfall_stubgen) which is generic over a "provider"/"implementation" type constrained to a generated trait. This trait would include all of the methods that currently contain a todo! item, with the addition that some values, such as the property_name, would be implemented as an enum for each function. The generated "adapter" generic type continues to be the type that implements Adapter as a trait, and it performs the current string-matching/unreachable! to find the value of that enum.

With this alternative model, each time the schema is changed, the generated module should be entirely regenerated, and the "provider"/"implementation" type would fail to compile due to differences in the trait or enums unless the new options are handled. With a concession to compile times (though possibly not a wise concession), this could even be implemented as a proc-macro so that the schema is updated every build.

I'm curious if this idea makes sense, or if it is fundamentally flawed in some way. I am very interested in hearing feedback from experienced users of this crate.

[obi1kenobi]

I'm not sure I was able to fully understand the vision you have, but I'd be happy to see a prototype if you wanted to put one together!

trustfall_stubgen definitely isn't the right tool for incrementally evolving a schema, and there's certainly room for alternative approaches. I'd be happy to see what you come up with!