Luiz Capitulino <lcapitul...@redhat.com> writes:
> On Wed, 18 Nov 2009 16:16:11 +0100
> Markus Armbruster <arm...@redhat.com> wrote:
>
>> Luiz Capitulino <lcapitul...@redhat.com> writes:
>
> [...]
>
>> > +static const char *append_field(QString *outstr, const QError *qerror,
>> > + const char *start)
>> > +{
>> > + QObject *obj;
>> > + QDict *qdict;
>> > + QString *key_qs;
>> > + const char *end, *key;
>> > +
>> > + if (*start != '%')
>> > + parse_error(qerror, '%');
>>
>> Can't happen, because it gets called only with *start == '%'. Taking
>> pointer to the character following the '%' as argument would sidestep
>> the issue. But I'm fine with leaving it as is.
>
> It's just an assertion.
It's not coded as an assertion. If we ever do coverage testing, it'll
stick out. But again, I'm fine with it.
>> > + start++;
>> > + if (*start != '(')
>> > + parse_error(qerror, '(');
>> > + start++;
>> > +
>> > + end = strchr(start, ')');
>> > + if (!end)
>> > + parse_error(qerror, ')');
>> > +
>> > + key_qs = qstring_from_substr(start, 0, end - start - 1);
>> > + key = qstring_get_str(key_qs);
>> > +
>> > + qdict = qobject_to_qdict(qdict_get(qerror->error, "data"));
>> > + obj = qdict_get(qdict, key);
>> > + if (!obj) {
>> > + qerror_abort(qerror, "key '%s' not found in QDict", key);
>> > + }
>> > +
>> > + switch (qobject_type(obj)) {
>> > + case QTYPE_QSTRING:
>> > + qstring_append(outstr, qdict_get_str(qdict, key));
>> > + break;
>> > + case QTYPE_QINT:
>> > + qstring_append_int(outstr, qdict_get_int(qdict, key));
>> > + break;
>> > + default:
>> > + qerror_abort(qerror, "invalid type '%c'", qobject_type(obj));
>> > + }
>> > +
>> > + QDECREF(key_qs);
>>
>> Looks like you create key_qs just because it's a convenient way to
>> extract key zero-terminated. Correct?
>
> Yes, as a substring of 'desc', which is passed through 'start'.
Funny that the convenient way to extract a substring is to go through
QString. Fine with me.
> [...]
>
>> > diff --git a/qjson.c b/qjson.c
>> > index 12e6cf0..60c904d 100644
>> > --- a/qjson.c
>> > +++ b/qjson.c
>> > @@ -224,6 +224,8 @@ static void to_json(const QObject *obj, QString *str)
>> > }
>> > break;
>> > }
>> > + case QTYPE_QERROR:
>> > + /* XXX: should QError be emitted? */
>>
>> Pros & cons?
>
> It's probably convenient to have qjson emitting QError, I'm unsure
> if we should do that for all kinds of QObjects though.
For a general purpose system, I'd recommend to cover all types. But as
long as this has just one user (QEMU), it can use the special purpose
excuse not to.
>> > case QTYPE_NONE:
>> > break;
>> > }
>> > diff --git a/qobject.h b/qobject.h
>> > index 2270ec1..07de211 100644
>> > --- a/qobject.h
>> > +++ b/qobject.h
>> > @@ -43,6 +43,7 @@ typedef enum {
>> > QTYPE_QLIST,
>> > QTYPE_QFLOAT,
>> > QTYPE_QBOOL,
>> > + QTYPE_QERROR,
>> > } qtype_code;
>> >
>> > struct QObject;
>>
>> Erroneous QERRs are detected only when they're passed to
>> qerror_from_info() at run-time, i.e. when the error happens. Likewise
>> for erroneous qerror_table[].desc. Perhaps a unit test to ensure
>> qerror_table[] is sane would make sense. Can't protect from passing
>> unknown errors to qerror_from_info(), but that shouldn't be a problem in
>> practice.
>
> We could also have a debug function that could run once at startup
> and do the check.
Yes.
