Re: Protocol ambiguity: want_reply vs CHANNEL_CLOSE

Matt Johnston <[email protected]> wrote:

> Given there are millions of deployed servers that won't change, would
> it be best to just publish a RFC and keep using a compatibility table?

In other words, designate one answer as 'right' and recommended for
new servers, but also keep a list of implementations known to do the
other thing? I suppose that's not too onerous; it's a bit out of the
ordinary as bug workarounds go, but then a fundamental protocol issue
that's been around forever _is_ a bit out of the ordinary.

> Out of interest has PuTTY encountered this problem in the wild?

Yes. We use a channel request with want_reply set as a means of
measuring round-trip time so as to tune our window size, which means
that if a server closes a channel at all it's quite common for us to
have just sent a channel request at the point it does so.

And yes, we have encountered at least one server with each policy!

Cheers,
Simon
-- 
Simon Tatham         "Every person has a thinking part that wonders what
<[email protected]>    the part that isn't thinking isn't thinking about."