Re: [Xen-devel] [PATCH V4 11/24] libxl/gentypes.py: include discriminator in JSON output

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

On Wed, 2014-05-07 at 11:35 +0100, Wei Liu wrote:
> On Wed, May 07, 2014 at 11:00:26AM +0100, Ian Campbell wrote:
> > On Tue, 2014-05-06 at 14:09 +0100, Wei Liu wrote:
> > > On Tue, May 06, 2014 at 02:03:49PM +0100, Ian Campbell wrote:
> > > > On Thu, 2014-05-01 at 13:58 +0100, Wei Liu wrote:
> > > > > Parser relies on the discriminator to go to correct branch.
> > > > > 
> > > > > Signed-off-by: Wei Liu <wei.liu2@xxxxxxxxxx>
> > > > 
> > > > Acked-by: Ian Campbell <ian.campbell@citrix>
> > > > 
> > > > But, an idea: What about calling the field which refers to the union by
> > > > it's discriminated name, i.e. "u.pv": { map of things } instead of
> > > > "type": "pv", "u" { map of things } ?
> > 
> > > But that breaks API, doesn't it?
> > 
> > Doh, yes.
> > 
> > > If we can safely assume that there's no
> > > existing user of this JSON API, changing it won't be hard.
> > 
> > Is it even possible to sanely use things as they are? The existence of
> > this patch would seem to suggest not?
> > 
> 
> There's actual a horrible way to hack around current implementation: if
> you calculate the intersection of those fields in "pv" and "hvm" you can
> actually discriminate things. Not sure if a sane person would do it
> though!

I think it is a pretty safe bet noone is doing this! (or at least they
probably have no illusions about it being sane!)

> However I would rather stay on the safe side.

I'm leaning towards we change this, although rather than "u.pv" as I
suggested earlier I think "type.pv" (so "discriminator field name"."enum
name") would be better.

Ian.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel