Re: [hybi] With deflate-stream, Close frame doesn't work as an end of data marker

[Takeshi Yoshino <[e-mail]>]

On Fri, Feb 18, 2011 at 09:33, <[e-mail]> wrote:

>
> Hi Takeshi,
>
> Good point. There is indeed a legitmate case where trailing bytes is
> present after
> Close frame.
>

It is important to note that after endpoint sent and received Close frame,
> it is considered
> that WebSocket layer close handshake has been complete ('wasClear' gets set
> to 1,
> etc.).
>
> What Takeshi pointed out is a good motivation for WS-layer to perform
> graceful
> shutdown at TCP layer as well. After Close frame has been exchanged
> successfully,
> the endpoint must start shutting down the local TCP connection with
> shutdown(SHUT_WR), and discard the trailing bytes, if any received, and
> then finally
> close the socket on receipt of EoS.
>

This works for any trailing bytes that could be caused by artifact of
> another deflate
> algorithm that may be introduced in the future, or just by a bug.
>
>
For this case, we can take another way without depending on TCP FIN
- have deflater compress Closing frame in BFINAL block (flush with Z_FINISH)
- have inflater drain compressed stream until it see EOB

This might work for most environments, but not all library/wrapper of zlib
expose API to recognize EOB/BFINAL immediately. For example, zlib module of
Python doesn't. See the explanation of zlib.Decompress.unused_data. If we
move Close frame out of compressed stream, we can use this, but it
complicates protocol.

I agree that using SHUT_WR + EoS is nice on platforms where they're
available, but I still want to provide a way that deals with this issue
without depending on them if possible.

Takeshi