Re: [libvirt PATCH 9/9] rpc: use new virt-nc binary for remote tunnelling

On Mon, 2020-07-20 at 18:21 +0100, Daniel P. Berrangé wrote: > On Mon, Jul 20, 2020 at 07:18:55PM +0200, Andrea Bolognani wrote: > > The other day I randomly realized the ssh-based transports already > > accept a 'netcat' URI parameter which can be used to point libvirt > > to a non-standard nc stand-in. With that in mind, is it really > > necessary to introduce another URI parameter? Can't we just reuse > > the existing one? > > Any binary provided there needs to implement netcat syntax. The whole > point of this work is that we don't want to use the netcat syntax since > it is impossible to know the correct socket path. We could special-case binaries called 'virt-nc' and use our internal syntax for those. Having two separate URI parameters to control the same knob sounds like trouble, especially since you can mix and match: if you try to connect to qemu+ssh://host/system?proxy=virt-nc&netcat=my-cool-nc for example, what happens? As far as I can tell virt-nc will be used, but it's certainly not as obvious as it would be if everything was controlled by a single URI parameter.

Actually, and I might be very well missing something because I looked at the code rather quickly, from what I can tell the default value for proxy will cause libvirt to always prefer virt-nc when available, which means that the URI qemu+ssh://host/system?netcat=my-cool-nc will suddenly stop using my-cool-nc and start using virt-nc after libvirt has been upgraded - a breaking change.