generate schema from models?

[hampsterx]

is there any way to generate a data dict of the model structure (field + type)? This would allow to generate typescript definitions for example ala poor mans openAPI generator type tool?

[Fatal1ty]

Hi. There is get_type_hints function you can use for this.

[hampsterx]

ahh thats true, yeah I had a quick look into your code afterwards and realised it should be quite straightforward. Cheers~

[adriangb]

Are there any plans for a non "poor mans" version? Or is there a compatible library that can generate jsonschema from a data class?
In the same vein, is there any plan to express types like {"type": "string", "pattern": "someregex"} which is a part of jsonschema?
Thanks!

[Fatal1ty]

Are there any plans for a non "poor mans" version? Or is there a compatible library that can generate jsonschema from a data class?

Unfortunately, I don't know of any compatible library but you could try to reuse field metadata to generate a simple json schema.

In the same vein, is there any plan to express types like {"type": "string", "pattern": "someregex"} which is a part of jsonschema?

To do this we need extensible validation mechanism which is not ready yet.

[hampsterx]

sorry only tangently related but..

To do this we need extensible validation mechanism which is not ready yet.

Any plans to support late initialization?

eg

foo = Bar()
foo.bar = 123
...
# validate..

This along with schema would be perfect. In our situation we have quite large response models that are built up and would be nicer than passing the full dict to the constructor.

[adriangb]

I do agree that separating construction from validation would be best (vs. pydantic where the constructor validates by default and you have to use construct to build a model without validation).

[Fatal1ty]

In our situation we have quite large response models that are built up and would be nicer than passing the full dict to the constructor.

@hampsterx How do you build that models? If you want to deserialize dataclasses with another library and validate with mashumaro I wouldn’t say it’s a good idea.

I do agree that separating construction from validation would be best (vs. pydantic where the constructor validates by default and you have to use construct to build a model without validation).

@adriangb If I read you right, you want to be able to choose either to validate or not during the deserialization process. Or do you want to trigger validation manually at any time after an instance is created?

[hampsterx]

Currently we just build up a big response dict but want to use models so we can lock it down and ideally generate client libraries (eg typescript, dart, etc) from the (json) schema. would be much nicer to start with Foo() and then build it up as we go rather than Foo(ret) at the end.

[adriangb]

I think manually triggering validation is more valuable:

1. It encompasses a the former approach since running validation immediately after construction is equivalent to validating during construction.
2. It allows more flexibility, e.g. the builder pattern shown above.

[Fatal1ty]

Are there any plans for a non "poor mans" version? Or is there a compatible library that can generate jsonschema from a data class?
In the same vein, is there any plan to express types like {"type": "string", "pattern": "someregex"} which is a part of jsonschema?
Thanks!

Since mashumaro has grown and become quite stable I’ve revised my opinion about jsonschema generation. This is what I’m currently working on as part of the following issue:

• JSON Schema generation #98

I don’t have a lot of experience in using jsonschema in the wild, so any help and suggestions are very welcome!

[Fatal1ty]

Well, the pull request is merged, documentation is here. Everyone is welcome to give it a try 🎉