0.26.0 (2025-08-26)

Breaking Changes

Change some union variant names

When creating a union with oneOf, anyOf, or a list of type, the name of each variant used to be type_{index}
where the index is based on the order of the types in the union.

This made some modules difficult to understand, what is a my_type_type_0 after all?
It also meant that reordering union members, while not a breaking change to the API, would be a breaking change
for generated clients.

Now, if an individual variant has a title attribute, that title will be used in the name instead.
This is only an enhancement for documents which use title in union variants, and only a breaking change for
inline models (not #/components/schemas which should already have used more descriptive names).

Thanks @wallagib for PR #962!

Features

Support patterned and default HTTP statuses

HTTP statuses like 2XX and default are now supported!

A big thank you to:

• @PSU3D0 for PR #973 (eons ago 😅)
• @obs-gh-peterkolloch for PR #1300
• @goodsonjr for PR #1304

Closes #1271 and #832

Note

Custom template users: the endpoint.responses type has changed quite a bit. Check out #1303 for the changes.

0.25.3 (2025-07-21)

Features

• Add --meta uv for generating astral-sh/uv compatible packages. (#1286)
• Switch to uv_build build backend. (#1290)

0.25.2 (2025-07-03)

Fixes

• Import error for types.FileType (#1274) (#1278)

0.25.1 (2025-06-19)

Fixes

• Support ruff 0.12 (#1270)

0.25.0 (2025-06-06)

Breaking Changes

• Raise minimum httpx version to 0.23

Removed ability to set an array as a multipart body

Previously, when defining a request's body as multipart/form-data, the generator would attempt to generate code
for both object schemas and array schemas. However, most arrays could not generate valid multipart bodies, as
there would be no field names (required to set the Content-Disposition headers).

The code to generate any body for multipart/form-data where the schema is array has been removed, and any such
bodies will be skipped. This is not expected to be a breaking change in practice, since the code generated would
probably never work.

If you have a use-case for multipart/form-data with an array schema, please open a new discussion with an example schema and the desired functional Python code.

Change default multipart array serialization

Previously, any arrays of values in a multipart/form-data body would be serialized as an application/json part.
This matches the default behavior specified by OpenAPI and supports arrays of files (binary format strings).
However, because this generator doesn't yet support specifying encoding per property, this may result in
now-incorrect code when the encoding was explicitly set to application/json for arrays of scalar values.

PR #938 fixes #692. Thanks @micha91 for the fix, @ratgen and @FabianSchurig for testing, and @davidlizeng for the original report... many years ago 😅.

0.24.3 (2025-03-31)

Features

Adding support for named integer enums

#1214 by @barrybarrette

Adding support for named integer enums via an optional extension, x-enum-varnames.

This extension is added to the schema inline with the enum definition:

"MyEnum": {
    "enum": [
        0,
        1,
        2,
        3,
        4,
        5,
        6,
        99
    ],
    "type": "integer",
    "format": "int32",
    "x-enum-varnames": [
        "Deinstalled",
        "Installed",
        "Upcoming_Site",
        "Lab_Site",
        "Pending_Deinstall",
        "Suspended",
        "Install_In_Progress",
        "Unknown"
    ]
}

The result:

0.24.2 (2025-03-22)

Fixes

Make lists of models and enums work correctly in custom templates

Lists of model and enum classes should be available to custom templates via the Jinja
variables openapi.models and openapi.enums, but these were being passed in a way that made
them always appear empty. This has been fixed so a custom template can now iterate over them.

Closes #1188.

0.24.1 (2025-03-15)

Features

• allow Ruff to 0.10 (#1220)
• allow Ruff 0.11 (#1222)
• Allow any Mapping in generated from_dict functions (#1211)

Fixes

Always parse $ref as a reference

If additional attributes were included with a $ref (for example title or description), the property could be
interpreted as a new type instead of a reference, usually resulting in Any in the generated code.
Now, any sibling properties to $ref will properly be ignored, as per the OpenAPI specification.

Thanks @nkrishnaswami!

0.24.0 (2025-03-03)

Breaking Changes

Support $ref in responses

Previously, using a $ref to define a response was ignored, the code to call the endpoint was still generated, but
the response would not be parsed. Now, responses defined with $ref will be used to generate the response model, which
will parse the response at runtime.

If a $ref is incorrect or uses a feature that is not supported by the generator, these endpoints will start failing to
generate.

Features

Make config available in custom templates

The configuration options object is now exposed as a variable called config in Jinja2 templates.

Add docstrings_on_attributes config setting

Setting this option to true changes the docstring behavior in model classes: for any attribute that have a non-empty description, instead of describing the attribute as part of the class's docstring, the description will appear in an individual docstring for that attribute.

0.23.1 (2025-01-13)

Features

• allow Ruff 0.9 (#1192)