Re: Asynchronous DNS

[Alain Schneble]

Lars Ingebrigtsen <address@hidden> writes:

> Alain Schneble <address@hidden> writes:
>
>> - Current (Lisp) API won't change at all.  No need to extend :nowait.
>>   Current semantics of nil and t remain.  We do not introduce any new
>>   idioms or configurations (e.g. that would be the case with 'immediate)
>>   that we might regret in the future.  It should be fully backward
>>   compatible.
>
> Well, process-send on a newly made make-network-connection :nowait t
> will no longer fail, which is an API change.  :-)

I categorized this as a minor non-breaking bugfix, for the good, not the
bad.  At least, that was my intention.  :-)

>> Did I miss something?
>
> The only thing missing is whether to have a new state for TLS
> connections.  When you've asked for a TLS connection, all these commands
> should block until TLS has been negotiated, too.  We can have the TLS
> commands check for that explicitly (which makes the blocking :nowait
> users happy), but we also need a way to tell the non-blocking :nowait
> users that the connection is now ready for non-blocking usage.
>
> We could just leave the connection in 'connecting state, perhaps...

FWIW, that sounds very reasonable from my bird's eye view.  I really
kind of ignored the HAVE_GNUTLS paths until now -- which probably was a
mistake...  I'll study them later today.  A solution where client code
is not exposed to a new process state would be the preferred one, I
think.