Re: [libmicrohttpd] suspend/resume patch

[José Bollo]

Le dimanche 29 mai 2016 à 20:53 +0200, Christian Grothoff a écrit :
> On 05/17/2016 09:53 AM, José Bollo wrote:
> > 
> > Le mercredi 20 avril 2016 à 20:41 +0300, Evgeny Grin a écrit :
> > Please could check again whether the following patch that avoid to
> > close the connection when upgrade is called is valid or not.
> > 
> > diff -Naur a/src/microhttpd/connection.c
> > b/src/microhttpd/connection.c
> > --- a/src/microhttpd/connection.c   2016-04-08
> > 19:02:26.000000000
> > +0000
> > +++ b/src/microhttpd/connection.c   2016-04-08
> > 19:02:26.000000000
> > +0000
> > @@ -731,8 +731,7 @@
> >    {
> >      if (NULL == end)
> >        return MHD_YES;
> > -    if ( (MHD_str_equal_caseless_ (end, "close")) ||
> > -         (MHD_str_equal_caseless_ (end, "upgrade")) )
> > +    if ( (MHD_str_equal_caseless_ (end, "close")) )
> >        return MHD_NO;
> >     return MHD_YES;
> >    }
> 
> Well, as usual the situation isn't quite that simple. As you can see
> in
> 
> http://lists.gnu.org/archive/html/libmicrohttpd/2014-12/msg00000.html
> 
> this case was actually introduced to help Gregory support "upgrade",
> so
> now you're asking us to remove to help you support "upgrade".
> Funny...

Hi Christian,

I let time working and ideas well up but now I come back on the
question starved of more answer.

>From the reading of the thread, the situation isn't clear. Gregory
wrote that it wasn't enough.

I am skeptical and believe that no upgrading implementation can be
implemented when the socket is closed. I tried to clone the descriptor
but it is not solving the issue because the connection is closed.

I suspect that Gregory Junker has no working implementation.

Gregory, if still here, can you comment?

Is any other people trying to use upgrade?

Best regards
José Bollo

> 
> As discussed in the old threat, the issue is simply more complicated
> and
> neither version will fully do; "upgrade" needs more logic in various
> places, as we need to not generate "Connection: Keep-alive" (as
> Gregory
> points out), but also not close the socket.
> 
>