Re: Splitting transport and application error code spaces

[Mark Nottingham <[e-mail]>]

Personally -- I'm +1 on this. Having a clear distinction is best.


> On 10 Aug 2017, at 10:40 pm, Martin Thomson <[e-mail]> wrote:
> 
> A while back we decided to use a single error code space.  But in
> https://github.com/quicwg/base-drafts/issues/485, I noticed that we
> have an implicit requirement in the protocol not to have the transport
> close streams.  If the transport resets streams, it could destroy
> critical application state.
> 
> The neatest way to enforce the separation of application and transport
> is to create an application error space:
> 
>  https://github.com/quicwg/base-drafts/pull/722
> 
> This splits the error codes for application protocols from the
> transport codes.  To do this, it splits CONNECTION_CLOSE into
> TRANSPORT_CLOSE and APPLICATION_CLOSE.  These two frames have
> identical format and semantics, but use different error code spaces.
> RST_STREAM and STOP_SENDING now only carry application error codes,
> making it clear that resetting streams is the domain of the
> application protocol.
> 
> In doing this, I noticed that the error code space is ludicrously
> large.  So I have a companion PR that shrinks it to a much more
> manageable 16 bits:
> 
>  https://github.com/quicwg/base-drafts/pull/723
> 

--
Mark Nottingham   https://www.mnot.net/