> -----Original Message----- > From: Anthony PERARD [mailto:anthony.per...@citrix.com] > Sent: 19 December 2018 14:01 > To: Paul Durrant <paul.durr...@citrix.com> > Cc: qemu-de...@nongnu.org; qemu-block@nongnu.org; xen- > de...@lists.xenproject.org; Kevin Wolf <kw...@redhat.com>; Max Reitz > <mre...@redhat.com>; Stefano Stabellini <sstabell...@kernel.org> > Subject: Re: [PATCH v6 16/18] xen: automatically create XenBlockDevice-s > > On Wed, Dec 19, 2018 at 12:43:24PM +0000, Paul Durrant wrote: > > > Kevin seems to say that this could be done without the _flat_confused > > > version. The flat_confused version seems to be useful just because > > > the key "cache.direct" is used earlier, and because everything in > qdict > > > is a string. > > > > It could be, but there's a good reason for wanting everything as a > string, and that is so that I can do a qdict_iter to generate my trace > message. Also I really don't want to get too elaborate here... this is > supposed to be mimicking what would normally come via a json blob, and > that would start out as strings. > > Mimic JSON ? Not really. JSON has types. If the toolstack wanted > cache.direct or read-only option on a blockdev, it will need to use the > bool type as string type will be rejected. The expected types when > issuing a QMP request can be found in "qapi/block-core.json", for the > command "blockdev-add". > > Also, there is a comment on the qobject_input_visitor_new_flat_confused > function, it reads: > The block subsystem uses this function to visit its flat QDict with > possibly confused scalar types. It should not be used for anything > else, and it should go away once the block subsystem has been > cleaned up. > > We might as well avoid using it right now, as it's easy to do so.
Ah, I'd not noticed that comment. In that case, yes it should be avoided. Paul > > -- > Anthony PERARD