Re: [Qemu-devel] Patch to improve handling of server sockets

[Gerd Hoffmann <kraxel@redhat.com>]

   Hi,

>> You'll run into qmp for sure when forward-porting the patches to the
>> latest qemu bits. It is the machine-readable version of the monitor
>> protocol (in qemu 0.12+).
>
> I guess that's the qemu_opt_set() calls at the end of inet_listen_opts()?

See docs in QMP/*, the changes in monitor.c and q${type}.[ch]

qemu_opt_set() in inet_listen_opts() is only slightly related.  It is 
used to report back the address we've actually bound to.  Used by 'info 
chardev' and I think vnc too.  Yes, that has to be changed somehow ...

>> Second we have to care about the current users (especially libvirt).
>
> Wouldn't the users of that bit of information run it through
> getaddrinfo() anyways when trying to connect? So to them it shouldn't
> matter whether the name or an ASCII representation of the address is used.

I don't know how it is used.

>> Today qemu usually reports the address I think. Thus I tend to stick
>> to addresses to keep them happy.
>
> But wouldn't going from single address to multiple addresses be a bigger
> change for the users (and likely break them all) while going from
> address to name would only break those that were not using getaddrinfo()
> to translate the address into its binary representation.

It is probably best to bring this up on the libvirt list, this is the
most important user of those interfaces and I think other virtual machine
management folks are reading there too.

I personally don't care too much which way we pick.

> OTOH, going for multiple addresses would also allow starting qemu with
> more than a single -vnc option, which doesn't seem to be possible right
> now, and wich might come handy in situations when the set of addresses a
> qemu instance should be listening on is not available as a single DNS name.

Good point.

cheers,
   Gerd