CLI - support for multipart form data request body

[baywet]

details in #220 and linked pull requests.

[sebastienlevert]

Support for input file only in multipart scenarios

[KennethHoff]

Is this referring to the fact that multipart/form-data APIs uses a generic MultipartBody instead of a custom type-safe "MyCustomForm" model like how Json APIs function, or is this something else?

Around a year ago I tested out Kiota for TypeScript because I really liked it on my .Net application (which, at least at the time, interacted exclusively with JSON endpoints), but decided not to use it because I assumed it wasn't fully implemented yet. Earlier today I stumbled upon an external multipart/form-data endpoint on my .Net API as well, and I was equally surprised to see that MultipartBody was used in the C# client as well.

This begs the question; Is my schema incorrect and MultipartBody is a fallback "we were unable to parse the schema, so here's an escape-hatch", or are we supposed to interact with forms through this type-unsafe MultipartBody type? It feels so weird to me that Kiota generates a fairly good (Nullable notwithstanding..) type-safe models for JSON APIs, but falls back to some cryptic MultipartBody type for form APIs.

[baywet]

This issue is specific to generating CLI clients.

But to answer your question, multipart requires more information than "other serialization formats":

• each part needs a name
• each part needs a content type
• file parts might also have additional metadata like the file name etc...

This is why kiota projects a reference to that wrapper object, so customers can assemble the payload from a mix of structured (generated types) and unstructured parts.

Let us know if you have any additional comments or questions.

[KennethHoff]

Could you not, at the very least, generate a stub like the following:

{
   public MultipartFormDataContent FirstName { get; init; }
   public MultipartFormDataContent LastName { get; init; }
   public MultipartFormDataContent File { get; init; }
}

Such that you can at least know which pros exist

[baywet]

That's a good suggestion. However the parent type for those properties also needs to handle the boundary.