Re: [Qemu-devel] [PATCH 12/18] block-qdict: Clean up qdict_crumple() a bit

[Markus Armbruster]

Daniel P. Berrangé <address@hidden> writes:

> On Thu, Jun 14, 2018 at 10:40:58AM +0200, Kevin Wolf wrote:
>> Am 13.06.2018 um 17:23 hat Markus Armbruster geschrieben:
>> > Kevin Wolf <address@hidden> writes:
>> > 
>> > > Am 12.06.2018 um 14:58 hat Markus Armbruster geschrieben:
>> > >> When you mix scalar and non-scalar keys, whether you get an "already
>> > >> set as scalar" or an "already set as dict" error depends on qdict
>> > >> iteration order.  Neither message makes much sense.  Replace by
>> > >> ""Cannot mix scalar and non-scalar keys".  This is similar to the
>> > >> message we get for mixing list and non-list keys.
>> > >> 
>> > >> I find qdict_crumple()'s first loop hard to understand.  Rearrange it
>> > >> and add a comment.
>> > >> 
>> > >> Signed-off-by: Markus Armbruster <address@hidden>
>> > >
>> > > To be honest, I found the old version of the loop more obvious.
>> > >
>> > >>  qobject/block-qdict.c | 42 +++++++++++++++++++++---------------------
>> > >>  1 file changed, 21 insertions(+), 21 deletions(-)
>> > >> 
>> > >> diff --git a/qobject/block-qdict.c b/qobject/block-qdict.c
>> > >> index a4e1c8d08f..35e9052816 100644
>> > >> --- a/qobject/block-qdict.c
>> > >> +++ b/qobject/block-qdict.c
>> > >> @@ -403,7 +403,7 @@ static int qdict_is_list(QDict *maybe_list, Error 
>> > >> **errp)
>> > >>  QObject *qdict_crumple(const QDict *src, Error **errp)
>> > >>  {
>> > >>      const QDictEntry *ent;
>> > >> -    QDict *two_level, *multi_level = NULL;
>> > >> +    QDict *two_level, *multi_level = NULL, *child_dict;
>> > >>      QObject *dst = NULL, *child;
>> > >>      size_t i;
>> > >>      char *prefix = NULL;
>> > >> @@ -422,29 +422,29 @@ QObject *qdict_crumple(const QDict *src, Error 
>> > >> **errp)
>> > >>          }
>> > >>  
>> > >>          qdict_split_flat_key(ent->key, &prefix, &suffix);
>> > >> -
>> > >>          child = qdict_get(two_level, prefix);
>> > >> +        child_dict = qobject_to(QDict, child);
>> > >> +
>> > >> +        if (child) {
>> > >> +            /*
>> > >> +             * An existing child must be a dict and @ent must be a
>> > >> +             * dict member (i.e. suffix not null), or else @ent
>> > >> +             * clashes.
>> > >> +             */
>> > >> +            if (!child_dict || !suffix) {
>> > >> +                error_setg(errp,
>> > >> +                           "Cannot mix scalar and non-scalar keys");
>> > >> +                goto error;
>> > >> +            }
>> > >> +        } else if (suffix) {
>> > >> +            child_dict = qdict_new();
>> > >> +            qdict_put(two_level, prefix, child_dict);
>> > >> +        } else {
>> > >> +            qdict_put_obj(two_level, prefix, qobject_ref(ent->value));
>> > >> +        }
>> > >
>> > > At least, can you please move the else branch to the if below so that
>> > > the addition of the new entry is symmetrical for both scalars and dicts?
>> > > As the code is, it mixes the conflict check, creation of the child dict
>> > > and addition of scalars (but not to the child dict) in a weird way in a
>> > > single if block.
>> > >
>> > > Or maybe organise it like this:
>> > >
>> > > if (child && !(child_dict && suffix)) {
>> > >     error
>> > > }
>> > >
>> > > if (suffix) {
>> > >     if (!child_dict) {
>> > >         create it
>> > >         add it to two_level
>> > >     }
>> > >     add entry to child_dict
>> > > } else {
>> > >     add entry to two_level
>> > > }
>> > 
>> > Fleshing out...
>> > 
>> >         if (child && !child_dict) {
>> >             /*
>> >              * @prefix already exists and it's a non-dictionary,
>> >              * i.e. we've seen a scalar with key @prefix.  The same
>> >              * key can't occur twice, therefore suffix must be
>> >              * non-null.
>> >              */
>> >             assert(suffix);
>> >             /*
>> >              * @ent has key @address@hidden, but we've already seen
>> >              * key @prefix: clash.
>> >              */
>> >             error_setg(errp, "Cannot mix scalar and non-scalar keys");
>> >             goto error;
>> >         }
>> 
>> This catches "foo.bar" after "foo", but not "foo" after "foo.bar".
>> 
>> >         if (suffix) {
>> >             if (!child_dict) {
>> >                 child_dict = qdict_new();
>> >                 qdict_put(two_level, prefix, child_dict);
>> >             }
>> >             qdict_put_obj(child_dict, suffix, qobject_ref(ent->value));
>> >         } else {
>> >             assert(!child);
>> 
>> So "foo" after "foo.bar" would fail this assertion.
>
> We got coverage of bad inputs in tests/check-qdict.c
>
> If the qdict_crumple_test_bad_inputs doesn't already cover these scenarios
> we should extend it to make sure we detect them during testing

Depends on qdict_first() / qdict_next() traversal order, which makes it
awkward.