Re: Upgrade status for impl draft 1

++

Always send the magic, at least until we have data that shows it is
detrimental.
If you don't get the magic, it isn't draft-http2/1, so unless you know what
it is, close the connection.

server->client magic. Dunno.


On Thu, Feb 21, 2013 at 11:02 PM, Willy Tarreau <w@1wt.eu> wrote:

> On Fri, Feb 22, 2013 at 03:51:13PM +1100, Mark Nottingham wrote:
> >
> > On 22/02/2013, at 3:50 AM, Martin Thomson <martin.thomson@gmail.com>
> wrote:
> >
> > > I took some different conclusions away:
> > >
> > > Specifically, I believe that we discussed having magic always
> > > regardless of how we got started, so that there was only one code
> > > path.  That wasn't firm, but I distinctly remember the conversation
> > > that lead to that conclusion.
> >
> > Works for me; I'm more interested in just getting something concrete
> written down.
> >
> > Anyone have a problem with that?
>
> I agree with Martin here that we should avoid different code paths as
> much as possible.
>
> > One thing we need to discuss is how servers should handle it when the
> magic
> > isn't sent, or isn't sent correctly; hard close?
>
> In my opinion it will simply be an invalid protocol talked over the
> wire if it does not match what is expected. We must be a bit stricter
> with protocol elements than what we used to do since HTTP/0.9.
>
> > Also, we haven't concluded on server->client magic. Do we have a real
> need
> > for it?
>
> Depends whether an unexpected server response might be parsed as a valid
> response or not. We could start with and remove it later if unneeded.
>
> Willy
>
>
>