Maximum recursion level reached in Python 3

[horejsek]

ujson fails with maximum level reached for numpy array of ints in Python 3. It's probably not problem of ujson, but anyway I think ujson should end with better message. Something like „xy is not serializable because it fails with“ so developer know what to look for.

➜  ~ python2
Python 2.7.9 
>>> import ujson
>>> import numpy
>>> ujson.dumps(numpy.array([1.0, 2.0, 3.0]))
'[1.0,2.0,3.0]'
>>> ujson.dumps(numpy.array([1, 2, 3]))
'[1,2,3]'

➜  ~ python3
Python 3.5.0
>>> import ujson
>>> import numpy
>>> ujson.dumps(numpy.array([1.0, 2.0, 3.0]))
'[1.0,2.0,3.0]'
>>> ujson.dumps(numpy.array([1, 2, 3]))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
OverflowError: Maximum recursion level reached

[rafaelreuber]

I'm facing a similar problem

[mthh]

EDIT: My bad, it wasn't fixing anything (I just ran my test only on a numpy array of float which doesn't seems to be a problem (python 2 or 3) as it only occurs with arrays of integers on python 3);

I also saw this issue is related to #204 and #164. I guess it could be due to numpy.int64 which isn't instance of int in python 3 (while it is in python 2, and while numpy.float64 are instance of float in both cases).

It might be fixed by using the Number Protocol of the python API when checking value type but some people more familiar with python c api might need to confirm.

[horejsek]

Good work!

[tlbtlbtlb]

Also in python 2, arrays of numpy.float32 fail while float64 is fine:

>>> import numpy, ujson
>>> x64=numpy.array([1.0, 2.0])
>>> x32=numpy.array([1.0, 2.0], dtype=numpy.float32)
>>> ujson.dumps(x64)
'[1.0,2.0]'
>>> ujson.dumps(x32)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
OverflowError: Maximum recursion level reached

[Jahaja]

Resolved in 53f85b1 by no longer trying to serialize these.

[ringsaturn]

NumPy int64 has been fixed with caiyunapp/ultrajson with calling NumPy's header in C.