On 04/26/2016 04:33 PM, Cole Robinson wrote: > On 04/26/2016 04:26 PM, Laine Stump wrote: >> On 04/26/2016 03:21 PM, Cole Robinson wrote: >>> An embedded '/' in object names doesn't really work for any >>> of our stateful drivers, so let's explicitly reject it. >>> >>> https://bugzilla.redhat.com/show_bug.cgi?id=639923 >>> https://bugzilla.redhat.com/show_bug.cgi?id=787604 >> >> Was there any discussion I may have missed about why we shouldn't escape >> characters like this in the name rather than just forbidding them? >> > > No discussion besides some old comments in those bugs. But how do you escape > '/' for a unix filename? It's not really possible... we'd probably need some > kind of URI escape sequence to make it work, which just opens up another world > of pain making sure that the filename we put on disk doesn't collide with any > legitimately named VM. Crazyness ensues > >> This doesn't prevent us doing that in the future if we want though, and it >> does provide a useful error message where there was previous semi-mystery, so >> ACK to the series (but do we really need the feature flag for domains? Seems >> like a bit of overkill.) > > It's possible that existing VMs in non-stateful drivers already have a '/' in > the name, and we shouldn't accidentally reject them. Googling indicates that > vmware allows / in .vmx names at least. It may not matter in practice but I > wanted to be conservative. > > Thanks for the review, I'll wait a couple days to see if anyone else wants to > comment before pushing couple days == until after the release - Cole -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list