Re: [OAUTH-WG] Device Code expiration and syntax

[Justin Richer <[e-mail]>]

> On Mar 11, 2017, at 2:54 PM, William Denniss <[e-mail]> wrote:
> 
> On Sat, Mar 11, 2017 at 11:10 AM, Justin Richer <[e-mail] <mailto:[e-mail]>> wrote:
> We’re implementing support for the device code draft and had a question on what the “expiration” of the code refers to. Obviously, once the code has expired it can no longer be used. But when should the expiration count from? Say I have a code that’s good for 60 seconds, do I start the timer as soon as I issue the code to the client? Do I reset the timer when the user approves the client, to another 60 seconds? Or does that 60 seconds count for the entire transaction?
> 
> My read on it is the latter-- one timeout for the entire lifetime of the code regardless of its current state, with no resets. But I didn’t find good guidance in the document itself.
> 
> It's the expiry of the user_code and device_code pair, at which point the device will need to start-over with a new device authorization request.  The device wouldn't *have* to start a timer, as they will get an error during polling:
> 
>    expired_token
>       The "device_code" has expired.  The client will need to make a new
>       Device Authorization Request.
> 
> We should add some guidelines around expiry behavior.

OK, so it really is one expiration for the whole thing. The device doesn’t need to care (and I’ll bet you right now that, just like with access tokens, the overwhelmingly vast majority of devices won’t care about expires_in), but the authorization server certainly does, and we wanted to know the right place to set the timers.

> 
> Secondly, I had a question about the “response_type” parameter to the device endpoint. This parameter is required and it has a single, required value, with no registry or other possibility of extension. What’s the point? If it’s for “parallelism”, I’ll note that this is *not* the authorization endpoint (as the user is not present) and such constraints need not apply here.
> 
> Good points here. At a guess, it bled in from the OAuth spec. If it's not needed, we should remove it.
> 

I’d vote for removal, I don’t see the point.

 — Justin