Paolo Bonzini <pbonz...@redhat.com> writes:
> On 04/02/2017 01:45, Markus Armbruster wrote:
>>>> -drive driver=qcow2,
>>>> file.driver=gluster,
>>>> .volume=testvol,
>>>> .path=/path/a.qcow2,
>>>> .debug=9,
>>>> file.server.0.type=tcp,
>>>> .host=1.2.3.4,
>>>> .port=24007,
>>>> file.server.1.type=unix,
>>>> .socket=/var/run/glusterd.socket
>>>>
>>>> Mind, I'm not at all sure this is a *good* idea. I suspect it's more
>>>> magic than it's worth.
>>> As someone who likes dot syntax very much, I don't like it. If you
>>> structure it like this, it's OK, but then you can just write the full
>>> prefix (which gets the point across just as well because I can quickly
>>> tell from a glance that it's the same prefix).
>>>
>>> OTOH, when joined into a single line it doesn't change much in terms of
>>> legibility, in my opinion.
>>
>> Thanks!
>
> Actually I think it does improve legibility.
>
> It doesn't improve writability though, as anecdotally proved by Markus's
> own mistake.
>
> I am a fan of the dot syntax too. It seems to be the most incremental
> solution, and it's still as expressive as JSON.
Noted.
> _However_ we could also extend -readconfig to support JSON, i.e. instead of
>
> [drive "abc"]
> file = "foo"
>
> it could support
>
> { 'drive': { 'file: 'foo' }, 'id': 'abc' }
>
> In other words [ would introduce key-value QemuOpts with dot syntax,
> while { would introduce JSON.
Yes, we should support config files in JSON syntax. Not sure mixing INI
and JSON syntax in the same file is a good idea, though.
