Error when jsonify query result

[jonahfang]

code:

from sanic import response
...
    u1 = await User.create(nickname='fantixd')
    u2 = await User.get(u1.id)

    return response.json(u2)

Error:

  File "/usr/local/lib/python3.6/site-packages/sanic/response.py", line 246, in json
    return HTTPResponse(json_dumps(body, **kwargs), headers=headers,
OverflowError: Maximum recursion level reached

[fantix]

This is because ujson does not support serializing a SQLAlchemy clause element:

In [1]: import sqlalchemy, ujson

In [2]: ujson.dumps(sqlalchemy.select([123]))
---------------------------------------------------------------------------
OverflowError                             Traceback (most recent call last)
<ipython-input-2-2ef38474e0b5> in <module>()
----> 1 ujson.dumps(sqlalchemy.select([123]))

OverflowError: Maximum recursion level reached

And u2.query is exactly one of such. To bypass this defect, use a custom JSON encoder. See also sanic-org/sanic#905

[jonahfang]

I wrote my jsonify method as:

def jsonify_gino_model(rec):
    if type(rec) == list:
        return [jsonify_gino_model(r) for r in rec]

    if hasattr(rec,'__table__'):
        columns = rec.__table__.columns
        d = {}
        for c in columns:
            fld = str(c).split('.')[1]
            d[fld] = getattr(rec,fld)
    else:
        #rec is plain obj
        d = rec

    return d

[fantix]

Add a Model.to_dict() method might be helpful?

[jonahfang]

How abount Model.jsonify() ?

[fantix]

Nah, that would normally be a application-specific behavior, and GINO cannot jsonify all types of columns in a way that is acceptable for all scenarios, DateTime for example.

[jonahfang]

You are right , Model.to_dict() is good!

[fantix]

to_dict it is then. 😈

[qdzzyb2014]

how about implement __iter__？ @fantix

[jonahfang]

My new jsonify_model method changed to:

def jsonify_model(rec):
    if type(rec) == list:
        return [jsonify_model(r) for r in rec]

    if hasattr(rec,'__table__'):
        columns = rec.__table__.columns
        d = {}
        for c in columns:
            fld = str(c).split('.')[1]
            v = getattr(rec,fld)
            if type(v) == datetime.datetime:
                d[fld] = str(v)
            else:
                d[fld] = v
    elif type(rec) == asyncpg.Record:
        ds = rec.items()
        d = {}
        for vd in ds:
            if type(vd[1]) == datetime.datetime:
                d[vd[0]] = str(vd[1])   #otherwise return time int
            else:
                d[vd[0]] = vd[1]
    else:
        #rec is plain object
        d = rec

    return d

It seems that asyncpg.Record can get data by .items() (it implements iter?)

[fantix]

@qdzzyb2014 the thing is, Model should be a generic base class for user-customized models. We'll never know how user may iterate their model objects for. So defining a default __iter__ is not a good idea IMO.

[fantix]

@jonahfang asyncpg.Record is both a tuple and a dict at the same time, it has __iter__ as tuple and items as dict. However as stated previously, I'd prefer Model stay as a plain Python object, leaving the "mixin" features for users - one can always mix in a dict interface or an iterable interface, through the declarative_base parameters. As for your jsonify, I'd suggest this:

    if isinstance(rec, db.Model):
        d = {}
        for k, v in rec.to_dict().items():
            if isinstance(v, datetime.datetime):
                d[fld] = v.strftime('YOUR_FORMAT_HERE')
            else:
                d[fld] = v

[jonahfang]

The following code is a bit ugly:

if isinstance(rec, db.Model):
...

I have to got 'db' first?

[fantix]

Well hasattr(rec, 'to_dict') works too, but db is usually a global variable as MetaData is.

[jonahfang]

Can I write as:

import sqlalchemy as sa
...
if isinstance(rec,sa.MetaData):

[fantix]

No you cannot.