Executing the same code using Win32COM does not produce an error either.

can you show minimal example and how you figured out that type is changed?

do examples on this page work for you?

https://github.com/pythonnet/pythonnet/tree/develop

Just saw the C# examples for numpy. But I am going the other way around. I am trying to pass the numpy float64 object as an argument to the clr package. To give some context, the package is a customized dbapi wrapper.

import clr
import numpy as np
clr.AddReference('Custom.DB.API')
from Custom.DB.API import Query()

q = Query('select :num from dual')
q.parameter('num',np.float(1)) #works
q.parameter('num',np.long(1)) #works
q.parameter('num',np.double(1)) #breaks

But the same example using Win32Com instead works without any issues.

@vmuriart I could replicate your problem:

import clr
from System import Array, Object
import numpy as np

arr3=Array[Object]([np.float(1),np.long(1)])
list(arr3)  # returns [1.0, 1]

But:

arr4=Array[Object]([np.float(1),np.double(1)])

returns:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-53-fee0ccf03f95> in <module>()
----> 1 arr4=Array[Object]([np.float(1),np.double(1)])

TypeError: value cannot be converted to Object

@tonyroberts how can this issue be debugged? is this failing in the weird IL code?

@denfromufa just search for the error in the code (it's in converter.cs) and it might be fairly clear what's going on just from looking at the code immediately before the error. Otherwise, set a breakpoint where the error is being set, attach the visual studio debugger to your python process and run the python code and work back from there.

In this case I imagine that the C# code is checking for an exact type match with 'float' which is failing for np.double. If that is the case, changing it to an 'isinstance' style check instead would fix it I would think.

Pardon my low level of understanding. I surveyed the code, and from what I think I understand, on runtime.cs you are creating c-type objects from the python library that correspond to Python Objects. Then the code compares incoming objects against this reference object to determine if its of the same type to then convert it?
The numpy double isn't identifying itself as a python standard object unless I explicitly cast it to change itself into one. I'm browsing through the numpy documentation to see what type of C object it identifies itself as.
dont know if this helps maybe.

to be more precise numpy double fails here:

return new IntPtr((void*)(*((ulong*)p + n)));

call stack

Python.Runtime.dll!Python.Runtime.Runtime.PyObject_TYPE(System.IntPtr op) Line 898  C#
Python.Runtime.dll!Python.Runtime.Runtime.PyFloat_Check(System.IntPtr ob) Line 1235 C#
Python.Runtime.dll!Python.Runtime.Converter.ToManagedValue(System.IntPtr value, System.Type obType, out object result, bool setError) Line 311  C#
Python.Runtime.dll!Python.Runtime.Converter.ToManaged(System.IntPtr value, System.Type type, out object result, bool setError) Line 220 C#

locals

        op  {483440320} System.IntPtr
        p   0x000000001cd0b6c0  void*
        n   1   int

auto

        (ulong*)p   0x000000001cd0b6c0  ulong*
        is32bit false   bool
        n   1   int
        p   0x000000001cd0b6c0  void*

to make numpy floats and integerswork with pythonnet, it is necessary to useisinstance()withnumber.Realandnumber.Integralrespectively. simply trying to compare withpython float and intdoes not work fornumpy` objects.

That sounds logical. I tried to look at how pywin32 handles the conversion, but their source isn't as navigation friendly. Have you tested to see if any of the other numpy objects break as well?

Yep, bunch of them break, np.float did not break because it maps to float in Python.

I did not find how pywin32 works, but this document shows mapping between com types and numpy types in comtypes package:

http://pythonhosted.org/comtypes/#numpy-interop

This seems to be fixed.