Generate server and client code from Swagger (OpenAPI 2.0) specification
Swagger-to generates server and client code from Swagger (OpenAPI 2.0) specification; written in Python 3.
We wanted a code generator that is 1) easy to write, maintain and extend and that 2) produces readable code.
To achieve this, we wrote the tool in Python which we found to be far superior for code generation compared with other languages such as Go and Java. Since the original Swagger specification can become quite complex, we introduce an intermediate representation layer after parsing the specification and before generating the code. This layer allowed us to substantially simplify the generation code since it reduced the impedance mismatch by operating on abstract constructs (such as Maps) instead of directly referring to “raw” Swagger constructs (such as additional properties of an object).
The underlying language (Python), readable generated code and the intermediate representation layer are the three main differentiators to other similar code generation projects.
|Typescript + Angular||x|
Due to the lack of time, we can not cover all of the Swagger specification. The current generators work well for all of our simple and not-so-simple use cases. We believe they can also cover most of the other people’s needs as well.
Here is a non-comprehensive list:
- anonymous objects. We do not support anonymous objects in the definitions. Please define all objects as top level definitions.
- parameters. Go server and Python client support query, body, path and header parameters. Elm and Typescript + Angular clients support only query, body and path parameters. Cookie parameters are not supported.
- default values. We do not support default values due to the impedance mismatch between JSON and the target languages.
- validation of responses. Responses from the server are not validated due to the complexity and the run-time overhead.
- type hierarchy. We simply stack all the properties from allOf together without introducing any type hierarchy. First, this is due to Go and Elm which do not embrace type hierarchies. For example, you could use structural typing in Go, but then every piece of data would have to be a pointer with an interface which is clumsy. Second, it is due to expediency, since stacking properties in the intermediate representation layer means we do not have to care about them in upper, language-specific layers.
- Create a virtual environment:
python3 -m venv venv3
- Activate it:
- Install swagger_to with pip:
pip3 install swagger_to
To generate code, you need to invoke one of the swagger_to_*.py scripts. If the generated code exists, you need to specify --force command-line argument in order to overwrite the existing files.
We use the tag name to designate the generate code (e.g., package name in the Go server or service name in the Python client code). See this example from the test cases.
To generate an Elm client from a Swagger specification at /some/path/swagger.yaml, invoke:
swagger_to_elm_client.py \ --swagger_path /some/path/swagger.yaml \ --outdir /some/elm/path/src/your-client-directory
The generated code will have the following structure in /some/elm/path/src/your-client-directory:
|Client.elm||Elm Client, containing Models, Encoders, Decoders and Http Requests.|
|elm-package.sample.json||The Elm JSON Package containing the libraries used in Client.elm.|
Two non-standard libraries are used in the Client:
- “NoRedInk/elm-decode-pipeline” is used to decode JSON objects in a more scalable way than the one supported by the elm-lang libraries; and
- “elm-community/json-extra” is needed to encode Dict variables.
To generate a Go server from a Swagger specification at /some/path/swagger.yaml, invoke:
swagger_to_go_server.py \ --swagger_path /some/path/swagger.yaml \ --outdir /some/go/path/src/your-server-package
The generated code will have the following structure in /some/go/path/src/your-server-package:
|types.go||Go type definitions|
|jsonschemas.go||JSON schemas used to validate the input (using https://github.com/xeipuuv/gojsonschema)|
|handler_impl.sample.go||Empty implementation of the handler|
All the types defined in the Swagger are translated to types.go. The routing and validation code around the endpoints is generated in jsonschemas.go and routes.go.
The handler interface is given in handler.go. You need to implement the handler logic yourself. You can use handler_impl.sample.go as a starting point. We usually just copy/paste it to handler_impl.go and ignore handler_impl.sample.go in our repositories (e.g., by adding it to .gitignore).
In face of Swagger (i.e. API) changes, our workflow includes regenerating the code and using a diff tool like meld to sync the “old” handler_impl.go with the new handler_impl.sample.go.
parameters. We decided to generate the code to extract the parameters only from queries, bodies and paths.
It seemed difficult to automatically generate the code to extract form data arguments which would cover all the edge cases (such as files and duplicate entries). We still generate the handler function, but leave it to the programmer to extract these arguments manually from the request.
Due to lack of time, we did not implement header and cookie parameters. Contributions for these features are highly welcome!
response. The auto-generated code does not check that the response conforms to the specification. We found such checks to be unnecessarily constraining and almost impossible to implement for all the use cases.
To generate a Python 3 client from a Swagger specification at /some/path/swagger.yaml, invoke:
swagger_to_py_client.py \ --swagger_path /some/path/swagger.yaml \ --outpath /some/py/path/your_client_module.py
The generated client uses requests library.
Since input checks need to be performed by the server anyhow, we decided not to keep the code generator simple and more maintainable by including only the rudimentary type checks on the inputs. Hence all the sophisticated checks such as string patterns or casting of a Python integer to int32 are deliberately excluded. Analogously, we also do not validate the output coming from the server.
If time ever permits, we would like to include both more fine-grained input and output validation. At the moment, we did not confront any problems in the development process.
To generate a Python client from a Swagger specification at /some/path/swagger.yaml, invoke:
swagger_to_ts_angular5_client.py \ --swagger_path /some/path/swagger.yaml \ --outpath /some/typescript/path/your_client.ts
The generated client uses Angular http library. For the same reasons as for Python client, no checks are performed neither on the input nor on the output.
We found it important to have a uniform documentation style across all the Swagger specifications in an organization. To that end, swagger_to includes an utility to automatically check the style such as casing of the definition names, property names, descriptions and verb moods (present tense instead of imperative).To check the compliance of a Swagger specification at /some/path/swagger.yaml to the Swagger style guides, invoke:
swagger_style.py \ --swagger_path /some/path/swagger.yaml
The following checks are performed:
- The Swagger name is in camel case, its description capitalized, and the base path starts with a slash.
- Top level type definitions are in capitalized camel case, and properties are in snake case.
- Endpoint paths, operation_id and parameter names are in camel case.
- All descriptions:
- start with a present tense verb, whose first letter is lower case;
- have no leading or trailing whitespaces, tabs or new lines;
- contain either one line, or three or more, in which case the second is empty;
- end with a period.
- Endpoint paths start with a slash, and the responses contain “200” and “default”.
The script call returns 0 in case of no violations found, 1 in case of failed checks or 2 in case of illegal usage.
You can find the following examples useful for development:
- Swagger API: https://github.com/Parquery/swagger-to/blob/master/tests/cases/go_server/general/swagger.yaml
- Go Server generated code: https://github.com/Parquery/swagger-to/tree/master/tests/cases/go_server/general
- Py Client generated code: https://github.com/Parquery/swagger-to/blob/master/tests/cases/py_client/general/client.py
- Elm client generated code: https://github.com/Parquery/swagger-to/blob/master/tests/cases/elm_client/general/Client.elm
- Check out the repository.
- In the repository root, create the virtual environment:
python3 -m venv venv3
- Activate the virtual environment:
- Install the development dependencies:
pip3 install -e .[dev]
- Run precommit.py to execute pre-commit checks locally.
We follow Semantic Versioning. The version X.Y.Z indicates:
- X is the major version (backward-incompatible),
- Y is the minor version (backward-compatible), and
- Z is the patch version (backward-compatible bug fix).
Release history Release notifications | RSS feed
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.