Constructor bug (re JSON)

[Jonathan Eunice]

Here is what appears to be a stuf constructor bug:

>>> j = '{"this": [ {"and": 1} , {"and":2}, {"and":3}] }'
>>> import json
>>> json.loads(j)
{u'this': [{u'and': 1}, {u'and': 2}, {u'and': 3}]}

So far so good. Python logical equivalence to JSON structure. How about with OrderedDict?

>>> from collections import OrderedDict
>>> json.loads(j, object_pairs_hook=OrderedDict)
OrderedDict([(u'this', [OrderedDict([(u'and', 1)]), OrderedDict([(u'and', 2)]), OrderedDict([(u'and', 3)])])])

Great. How about with stuf?

>>> from stuf import stuf
>>> json.loads(j, object_pairs_hook=stuf)
stuf(this=stuf(and=3))

Not so much. Somehow stuf collapses down three parallel mappings into a single mapping, and erases the intervening array.

I thought I might work around the problem by converting to dict first:

>>> stuffer = lambda x: stuf(dict(x))
>>> json.loads(j, object_pairs_hook=stuffer)

But that gives me the same thing as when stuf is the hook itself. (Just for completeness, dict as an explicit hook works fine, like the default case.)

[Jonathan Eunice]

I did work around it with a factory function:

def stufize(d):
    if isinstance(d, list):
        return [ stufize(item) for item in d ]
    elif isinstance(d, dict):
        result = stuf()
        for k,v in d.items():
            result[k] = stufize(v)
        return result
    else:
        return d

But that seems a little grotesque.

[Jonathan Eunice]

I was not able to make an on-the-fly workaround. Probably because when it's boiled down, stuf really doesn't like encoding lists of stuf objects that have the same keys:

With dissimilar sub-keys, no problem:

>>> stuf(all=[stuf(a=1),stuf(b=2), stuf(c=3)])
stuf(all=stuf(a=1, c=3, b=2))

But with similar sub-keys, BOOM! Breakage and wrong results.

>>> stuf(all=[stuf(a=1),stuf(a=2), stuf(a=3)])
stuf(all=stuf(a=3))