add-auth-http-custom-support-in-rlc

[qiaozha]

fixes #1870
This PR is to support for http custom authentication like:

@useAuth({
  type: AuthType.http,
  scheme: "SharedAccessKey",
})

And use header like Authorization: SharedAccessKey <put my key here> to do the authentication.

As discussed in Azure/azure-sdk-for-js#25683, we will not add a commn policy in azure core. Instead, we will add a policy in the codegen side for http custom cases.

Right now, I am adding a policy in the client factory function createClient like

  client.pipeline.addPolicy({
    name: "customHttpAuthPolicy",
    async sendRequest(request, next) {
      request.headers.set(
        "Authorization",
        "SharedAccessKey " + credentials.key
      );
      return next(request);
    },
  });

And I am using KeyCredential as the credentials type in this case

export default function createClient(
  credentials: KeyCredential,
  options: ClientOptions = {}
): AuthHttpCustomClient;

[qiaozha]

do some self-reviews.

[xirzec]

I think the approach is solid.

Do we have any SDKs today that use this kind of auth that we can test with?

I was also curious if we wanted to support NamedKeyCredential as well though maybe that's too difficult to generically support at the moment.

[qiaozha]

Do we have any SDKs today that use this kind of auth that we can test with?

@xirzec I think event grid is using this way https://github.com/Azure/azure-rest-api-specs/blob/main/specification/eventgrid/Azure.Messaging.EventGrid/main.tsp#L6

[qiaozha]

Do we have any SDKs today that use this kind of auth that we can test with?

@xirzec I think event grid is using this way https://github.com/Azure/azure-rest-api-specs/blob/main/specification/eventgrid/Azure.Messaging.EventGrid/main.tsp#L6

Sorry, this is not a good example, eventgrid is using standard api key credential. I am not sure if it's their workaround or I am having some misunderstanding here. Confirming with @lmazuel offline.

[joheredi]

Looks good! Thanks @qiaozha for driving the conversations on this. I like the current approach

[MaryGao]

LGTM for the immediate problem and open with Option 3 once we notice this is a common requirement.

[qiaozha]

Confirmed with Laurent, that the eventgrid is not correctly written.