Re: [Qemu-devel] [PATCH 2/7] qapi: QMP input visitor, handle floats parsed as ints

[Michael Roth <mdroth@linux.vnet.ibm.com>]

On Fri, May 11, 2012 at 01:22:33PM -0300, Luiz Capitulino wrote:
> On Fri, 27 Apr 2012 15:21:18 -0500
> Michael Roth <mdroth@linux.vnet.ibm.com> wrote:
> 
> > JSON numbers can be interpreted as either integers or floating point
> > values depending on their representation. As a result, QMP input visitor
> > might visit a QInt when it was expecting a QFloat, so add handling to
> > account for this.
> > 
> > Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
> > ---
> >  qapi/qmp-input-visitor.c |    9 +++++++--
> >  1 files changed, 7 insertions(+), 2 deletions(-)
> > 
> > diff --git a/qapi/qmp-input-visitor.c b/qapi/qmp-input-visitor.c
> > index 4cdc47d..bc91134 100644
> > --- a/qapi/qmp-input-visitor.c
> > +++ b/qapi/qmp-input-visitor.c
> > @@ -246,13 +246,18 @@ static void qmp_input_type_number(Visitor *v, double *obj, const char *name,
> >      QmpInputVisitor *qiv = to_qiv(v);
> >      QObject *qobj = qmp_input_get_object(qiv, name);
> >  
> > -    if (!qobj || qobject_type(qobj) != QTYPE_QFLOAT) {
> > +    if (!qobj || (qobject_type(qobj) != QTYPE_QFLOAT &&
> > +        qobject_type(qobj) != QTYPE_QINT)) {
> >          error_set(errp, QERR_INVALID_PARAMETER_TYPE, name ? name : "null",
> >                    "double");
> 
> s/double/number
> 
> It's also important to note that migrate_set_downtime is (positively) affected
> by this change. Today, it only accepts a double, eg.:
> 
> { "execute": "migrate_set_downtime", "arguments": { "value": 1 } }
> {
>     "error": {
>         "class": "InvalidParameterType", 
>         "desc": "Invalid parameter type for 'value', expected: double", 
>         "data": {
>             "name": "value", 
>             "expected": "double"
>         }
>     }
> }
> 
> That's a bug (as it's documented to accept a number) and this patch fixes it.
> There's an interface change, but I think it won't cause problems in practice.
> 
> Please, fix the error message above and would be nice to get this (and patch
> 02/07) as a separate series for 1.1.

Hmm, this is kinda awkward because I don't think we can change it without
breaking any trees based off qom-next.

Since the error msg predates the bug fix (it just becomes more obviously
wrong as a result of the fix), can you pull these commits as is into your
QMP tree?

In the meantime I can send another patch, based on qom-next or qmp, that
fixes the error msg. We can then get all 3 into 1.1/master via QMP tree, and I
think qom-next should still merge cleanly back into master after 1.1

> 
> 
> >          return;
> >      }
> >  
> > -    *obj = qfloat_get_double(qobject_to_qfloat(qobj));
> > +    if (qobject_type(qobj) == QTYPE_QINT) {
> > +        *obj = qint_get_int(qobject_to_qint(qobj));
> > +    } else {
> > +        *obj = qfloat_get_double(qobject_to_qfloat(qobj));
> > +    }
> >  }
> >  
> >  static void qmp_input_start_optional(Visitor *v, bool *present,
>