fix: Fix multipart body file array

[micha91]

As described in #692 arrays of files are not handled correctly, if they are part of multipart/form-data. This is fixed in this PR by letting to_multipart return a List[Tuple[str, Any]] instead of a Dict[str, Any].

[dbanty]

Thanks for adding this support! I believe this is a breaking change, since it's possible someone was using the JSON serialization behavior before (even though it seems wrong in general). I'll need to do some manual testing just to make sure I know how to describe the breaking change 😅. I'd also love to get someone to test a real running API and verify it all functions (maybe you've already done this?)

[micha91]

Yes, I already use the updated model template as a custom template in a project. Unfortunately this project is not public yet and the API it is developed for, isn't publicly accessible, too

[ratgen]

Hi, just tested this out with an api I am working on. It works great.

[FabianSchurig]

@dbanty @micha91 Thank you so much for providing the openapi-python-client package and this PR!

I tested it in the following way:
pipx install git+https://github.com/openapi-generators/openapi-python-client.git@refs/pull/938/head

openapi-python-client update --path ../../swagger.json --config openapi-python-generator-config.yml

    "/upload": {
      "post": {
        "tags": [
          "Microsoft.KernelMemory.ServiceAssembly"
        ],
        "summary": "Upload a new document to the knowledge base",
        "description": "Upload a document consisting of one or more files to extract memories from. The extraction process happens asynchronously. If a document with the same ID already exists, it will be overwritten and the memories previously extracted will be updated.",
        "operationId": "UploadDocument",
        "requestBody": {
          "description": "Document to upload and extract memories from",
          "content": {
            "multipart/form-data": {
              "schema": {
                "type": "object",
                "properties": {
                  "index": {
                    "type": "string",
                    "description": "Name of the index where to store memories generated by the files."
                  },
                  "documentId": {
                    "type": "string",
                    "description": "Unique ID used for import pipeline and document ID."
                  },
                  "tags": {
                    "type": "object",
                    "additionalProperties": {
                      "type": "string"
                    },
                    "description": "Tags to apply to the memories extracted from the files."
                  },
                  "steps": {
                    "type": "array",
                    "items": {
                      "type": "string"
                    },
                    "description": "How to process the files, e.g. how to extract/chunk etc."
                  },
                  "files": {
                    "type": "array",
                    "items": {
                      "type": "string",
                      "format": "binary"
                    },
                    "description": "Files to process and extract memories from."
                  }
                }
              }
            }
          }
        },

In general it works!
I would greatly appreciate if we could have similar functionality to solve #692 in the main branch, soon.

@dbanty What do you think about an alternative where a user could just overwrite how a

"files": {
      "type": "array",
      "items": {
        "type": "string",
        "format": "binary"
      },

is handled?

This could either generate:

        files: Union[Unset, tuple[None, bytes, str]] = UNSET
        if not isinstance(self.files, Unset):
            _temp_files = []
            for files_item_data in self.files:
                files_item = files_item_data.to_tuple()

                _temp_files.append(files_item)
            files = (None, json.dumps(_temp_files).encode(), "application/json")

or

        files: Union[Unset, tuple[None, bytes, str]] = UNSET
        if not isinstance(self.files, Unset):
            files = files

The later works well in the case and should be just a minimal, non-breaking and optional change?