Received: from eggs.gnu.org ([2001:4830:134:3::10]:47735) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1YOrpM-0006v9-Sl for qemu-devel@nongnu.org; Fri, 20 Feb 2015 12:57:25 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1YOrpK-0000Zs-7A for qemu-devel@nongnu.org; Fri, 20 Feb 2015 12:57:24 -0500 Received: from mx1.redhat.com ([209.132.183.28]:52239) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1YOrpJ-0000Zi-Uh for qemu-devel@nongnu.org; Fri, 20 Feb 2015 12:57:22 -0500 From: Markus Armbruster References: <1424346029-5410-1-git-send-email-dgilbert@redhat.com> <1424346029-5410-3-git-send-email-dgilbert@redhat.com> <87r3tl9esl.fsf@blackfin.pond.sub.org> <54E75544.4090108@redhat.com> Date: Fri, 20 Feb 2015 18:57:16 +0100 In-Reply-To: <54E75544.4090108@redhat.com> (Eric Blake's message of "Fri, 20 Feb 2015 08:39:48 -0700") Message-ID: <871tlkzcsj.fsf@blackfin.pond.sub.org> MIME-Version: 1.0 Content-Type: text/plain Subject: Re: [Qemu-devel] [PATCH v3 2/3] Add migrate_incoming List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Eric Blake Cc: quintela@redhat.com, liang.z.li@intel.com, mjt@tls.msk.ru, qemu-devel@nongnu.org, "Dr. David Alan Gilbert (git)" , amit.shah@redhat.com, pbonzini@redhat.com Eric Blake writes: > On 02/20/2015 01:18 AM, Markus Armbruster wrote: >> I'd like Eric's opinion on on encoding configuration tuples as URIs >> rather than JSON in QMP. >> > >>> +{ 'command': 'migrate-incoming', 'data': {'uri': 'str' } } >>> + >>> # @xen-save-devices-state: >>> # >>> # Save the state of all devices to file. The RAM and the block devices >> >> Eric, what's your take on this? >> >> The general rule in QMP is "no ad hoc encoding of tuples in strings, use >> JSON objects instead". > > Yes, it would be nice to be type-safe, and have a QAPI union with a > discriminator of different URI types along with appropriate accompanying > data. But it is also new code (both in qemu to parse it, and in libvirt > to generate it), and we already have existing code that knows how to > generate the string-encoded tuple for the -incoming command line > argument that can be reused as-is with the new QMP command as proposed. > So even though it is less safe, I'm okay with this particular command > using an overloaded string argument. > > Down the road, if we WANT to add type-safety, we can make this command > take an anonymous union, where a 'str' value is the compact old style, > and where a dictionary value is the new discriminated union style. So > we aren't completely locked into non-type-safe string forever, but we > probably won't add a type-safe variant until (if) we ever add a new > migration format that just looks too ugly as a URI based on the > parameters it requires. I can accept this, but I want the rationale in a comment, so this exception to the QMP rule serves less well as bad example.