[http-client-csharp] cookie auth is generating incorrect code

[ArcturusZhang]

Our current code is treating cookie key authentication as if it is a header key authentication with the specific header name.
This is incorrect.

Proposal: since we could identify the type of authentication (including which kind of api key auth) from the output of TCGC, we should be able to report a diagnostic from the emitter and treat it as if there is no auth.

[JoshLove-msft]

@ArcturusZhang is your suggestion to handle this in the emitter as opposed to the generator? i.e. the emitter filters out the cookie header key from the authentication model?

[ArcturusZhang]

I think this is a thing that we must do in the emitter.
Our emitter is translating output from TCGC into ours, and TCGC already did this for us, they have this type defined as auth type:

export type HttpAuth = BasicAuth | BearerAuth | ApiKeyAuth<ApiKeyLocation, string> | Oauth2Auth<OAuth2Flow[]> | OpenIDConnectAuth | NoAuth;

We only support part of them in our emitter: ApiKeyAuth and Oauth2Auth.
The implementation in our MGC code always assume ApiKeyAuth is header, but actually the information from TCGC has a ApiKeyLocation and it could be header, query or cookie.
Related code here:

typespec/packages/http-client-csharp/emitter/src/lib/service-authentication.ts

Line 50 in 6ffff89

return { ApiKey: { Name: scheme.name } } as InputAuth;

[JoshLove-msft]

Should this be a cadl-ranch scenario? We should add it into the specs. We should warn and skip in the case of cookie auth.

[m-nash]

Created an issue for for the spector scenario #5584