Re: WGLC p1: Delimiting messages with multipart/byteranges

Question about trailing CRLF of a multipart body. RFC2046 defines

     multipart-body := [preamble CRLF]
                       dash-boundary transport-padding CRLF
                       body-part *encapsulation
                       close-delimiter transport-padding
                       [CRLF epilogue]


RFC2616 forbids "epilogue" in http messages, but what about the trailing
CRLF? I know browsers do include a trailing CRLF when submitting
multipart/form-data requests. Is that also the common practice for
multipart responses from servers?

Zhong Yu



On Mon, Apr 29, 2013 at 1:47 PM, Ben Niven-Jenkins
<ben@niven-jenkins.co.uk>wrote:

> Hi,
>
> The p1 HTTPbis draft has removed using multipart/byteranges as a message
> delimiter in 206 responses (which was previously allowed by RFC2616).
>
> I see this was originally tracked as Ticket #90 and the outcome was to
> deprecate such behaviour in all cases including in 206 responses with
> Content-Type: multipart/byteranges (which is the only case I care about).
>
> This decision makes previously conforming implementations (such as our
> reverse proxy implementation) now non-conformant and the alternatives of:
> (A) close connection after every multipart/byteranges response or (B)
> pre-calculate message body length (including header/boundary bytes for each
> byterange) in advance, are inconvenient.
>
> Would it be possible to re-allow use of multipart/byteranges as a message
> delimiter in the case where it is a 206 response AND there is no
> Transfer-Encoding AND there is no Content-Length so that implementations
> such as ours are classed as conformant again?
>
>
> Even if it is decided to stick with the current state of affairs in -22
> where generating 206 responses that use multipart/byteranges as a delimiter
> is forbidden, the specification should still specify how to parse such
> responses in a backwardly compatible manner as generating such responses
> was previously allowed & used and so User Agents should still expect to
> receive them from pre-httpbis implementations.
>
> Thanks
> Ben
>
>
>