On 03/07/2016 08:06 AM, Daniel P. Berrange wrote: > So looking at the current usage of qdict_flatten, qdict_extract_subqdict, > qdict_array_split and qdict_array_entries, it is basically only the block > layer. > > The root cause of this usage all stems from that fact that historically > the block layer was driven off QemuOpts CLI args which are a flat data > structure. Over time we've tried to bolt on recursive data structure > support, but the main APIs are still accepting QDicts whose contents are > flat. Increasingly I see the internal code is based on QAPI which is an > inherantly nested data structure, as a result the block layer is getting > more & more places where it has to process the flat data structure to > extract nested bits (qdict_extract_subqdict, qdict_array_entries and > qdict_array_split). Conversely when fed data coming from QMP it has to > flatten the data structure with qdict_flatten. > > With this new qdict_crumple() method (or equivalently you qdict_unflatten) > it strikes me we can significantly simplify the block layer to always use > a nested QDict internally. All that would be required is to call the > qdict_crumple() method in the places which have the flat QemuOpts derived > QDict. At that point we would potentially be able to delete all of > qdict_flatten, qdict_extract_subqdict, qdict_array_split and > qdict_array_entries > > Of course I'm not suggesting we do that for 2.6, but it actually doesn't > look like it would be that hard todo this conversion.
Agreed that it's too late for 2.6, but would make a good project for 2.7. For that matter, rather than passing a QDict around, I'd like to see if we could just directly use the QAPI types instead (the way you just recently converted from QDict to SocketAddress). -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature
