Newbie, sending pointers from python to C

[Prashant Saxena]

Hi,

Windows 7 x64
Python  2.7.11 x64
cffi 1.5.2
Microsoft python compiler for windows (vc++ 2008 x64)

I am trying my hands on cffi for the very first time. There is a C api in which most of functions has two parametes:

size_t arg_size, const void *arg_value Here we can pass either scalars or array. For this I have created a struct like this in cdef:

typedef struct {     void *data;     size_t size; } buffer; Now from python to pass data numpy array: p =  ffi.cast("float *", data.ctypes.data) buffer = ffi.new("buffer *", [p, data.nbytes]) lib.Test(buffer) Test function goes like this: void Test(buffer *data){ apiFunction(data->size, data->data); } This is working fine as far as I am concerned for numpy based arrays and python integers, but when I am passing float I am not able to receive it's value in C. printf("%f\n", data->data) actually prints 0.00000 but when I'm printing that float using: printf("%d\n", data->data) I am not receiving the decimal part. For e.g. if float is 462.236, it's only printing 462. I have no idea what's going on. In general I am looking for a solution that could work in given scenario. Cheers Prashant

[Damien Ruiz]

Hi Prashant,

You cannot print a float using %d in C, because %d only prints integer, and the integer part of 462.236 is 462.

What happens when you try to print 462.236 with %f?

[Prashant Saxena]

Hi Damien,

When I'm printing float, it's printing 0.00000. There is one more thing I came to know about numpy.
It's behaving inconsistently when using numpy.float32 data type. Yesterday night out of sudden numpy.float32
became C double and today morning It's working fine.

Right now the only solution I have is to use single element numpy array for scalars.

scalar = numpy.array(1, dtype=numpy.int32)

This works fine.

[Adam Gradzki]

Make sure your Numpy array is C-contiguous.

You can check this with ndarray flag: C_CONTIGUOUS

http://docs.scipy.org/doc/numpy-1.10.1/reference/generated/numpy.ndarray.flags.html

And convert it using:

http://docs.scipy.org/doc/numpy-1.10.0/reference/generated/numpy.ascontiguousarray.html

--
-- python-cffi: To unsubscribe from this group, send email to python-cffi...@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/python-cffi?hl=en
---
You received this message because you are subscribed to the Google Groups "python-cffi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to python-cffi...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Adam G]

Prashant,

I have some demo code using cffi, numpy, and float * you can take a look at for inspiration:

https://github.com/nszceta/cffi-numpy-demo/blob/master/imgsample.py

[Adam Gradzki]

Now that I am revisiting this code I am not sure if I have found a bug in CFFI or if it is in my own code.

git clone https://github.com/nszceta/cffi-numpy-demo.git

git checkout 8c27abe535554f5aac5bcc4e96ea9185b8a804bd

If you run `python3 -O imgsample.py` then the tests will complete without error.

However, the same code run without -O will fail! Unexplainable.

cffi==1.5.2, numpy==1.10.4, cc==Apple LLVM version 7.3.0 (clang-703.0.29)

Additionally, I found the behavior of ffi.from_buffer(my_input) to be buggy, and my code worked again when I replaced this piece of code with my_input.ctypes.data. 

Here is the diff that fixed things:

https://github.com/nszceta/cffi-numpy-demo/commit/3703e2489b4e548d2723cdffef8f7fd2ec93d3f3

[Armin Rigo]

Hi,

On 3 April 2016 at 08:31, Prashant Saxena <anima...@gmail.com> wrote:
>
>
> I am trying my hands on cffi for the very first time. There is a C api in which most of functions has two parametes:
>
> size_t arg_size,
> const void *arg_value

I am not sure to understand why, if you have a function which takes
these two arguments, you cannot directly call the C function with two
corresponding arguments. Why do you need the intermediate ``typedef
struct { void*; size_t;} buffer;``?

> Now from python to pass data numpy array:
> p = ffi.cast("float *", data.ctypes.data)
> buffer = ffi.new("buffer *", [p, data.nbytes])
> lib.Test(buffer)

You can do directly

lib.apiFunction(data.nbytes, ffi.cast("float *", data.ctypes.data))

> This is working fine as far as I am concerned for numpy based arrays and python integers,
> but when I am passing float I am not able to receive it's value in C.
> printf("%f\n", data->data)
> actually prints 0.00000 but when I'm printing that float using:
> printf("%d\n", data->data)
> I am not receiving the decimal part.

This sounds like a C problem. ``data->data`` is a pointer, right?
When you use ``%f`` to print a pointer, you get nonsense. When you
use ``%d`` to print a pointer, I would expect that you get nonsense as
well. I don't know why you get the integer part of some float number.
Are you sure you use exactly the line above, with the ``data->data``
field being a pointer?


A bientôt,

Armin.