Arrays are parsed incorrectly with multipart/form-data

[alfechner]

Description

I use multipart/form-data with a property bar which is of type array. The encoding is set to application/json. The spec looks like this:

openapi: 3.0.3
info:
  title: LIQUID Dataset Service
  version: 1.0.0
paths:
  /foo:
    post:
      x-openapi-router-controller: liquid_dataset_service.controllers.dataset_controller
      operationId: create_bar
      requestBody:
        required: true
        content:
          multipart/form-data:
            schema:
              type: object
              properties:
                bar:
                  type: array
                  items:
                    type: object
                    properties:
                      baz:
                        type: string
              required:
                - bar
            encoding:
              bar:
                contentType: application/json
      responses:
        204:
          description: No Content

Expected behaviour

When running the request I'd expect to get the bar property passed into the body argument of my controller function.

Actual behaviour

Server raises error: Validation error: '["{\\n \\"baz\\": \\"string\\"\\n}"]' is not of type 'object' - 'bar.0'.

To me it looks like that the string is not properly parsed if it's an array. To verify I changed the type of bar from array of object's to array of string's:

bar:
  type: array
  items:
    type: string

I passed two strings "one" and "two" and received this dict in my controller function: {'bar': ['["one"', '"two"]']}.

Steps to reproduce

Create a controller function, e.g.:

def create_bar(body):
    return

Adjust property x-openapi-router-controller accordingly.

Fire up the swagger page or use curl to run the request.

Additional info:

Output of the commands:

• python --version: Python 3.11.6
• pip show connexion | grep "^Version\:": Version: 3.0.5

[alfechner]

I found the resolve_form method invoked from

connexion/connexion/validators/form_data.py

Line 88 in f8f461c

data = self._uri_parser.resolve_form(form_data)

It's splitting the string parsed from the form data:

connexion/connexion/uri_parsing.py

Lines 163 to 166 in f8f461c

	if defn and defn["type"] == "array":
	form_data[k] = self._split(form_data[k], encoding, "form")
	elif "contentType" in encoding and all_json([encoding.get("contentType")]):
	form_data[k] = json.loads(form_data[k])

In my example form_data[k] is '["one","two"]' and self._split returns '["one"', '"two"]' instead of leaving it to the elif which correctly parses the JSON array.

I guess the splitting is place for URI params?

Shouldn't we always use json.loads(form_data[k]) if the encoding is set to application/json?

Maybe it's enough to switch the if with the elif and check for the content type first?

Would really appreciate feedback since our team is blocked by this 😅