/*
 * SPDX-FileCopyrightText: All Contributors to the PyTango project
 *
 * SPDX-License-Identifier: LGPL-3.0-or-later
 */

// This header file is just some template functions moved apart from
// device_attribute.cpp, and should only be included there.

#pragma once

#include "tango_numpy.h"

namespace PyDeviceAttribute
{

/// This callback is run to delete Tango::DevVarXArray* objects.
/// It is called by python. The array was associated with an attribute
/// value object that is not being used anymore.
/// @param ptr_ The array object.
/// @param type_ The type of the array objects. We need it to convert ptr_
///              to the proper type before deleting it. ex: Tango::DEV_SHORT.
#ifdef PYCAPSULE_OLD
template <long type>
static void _dev_var_x_array_deleter(void *ptr_)
{
    TANGO_DO_ON_ATTRIBUTE_DATA_TYPE_ID(type, delete static_cast<TANGO_const2arraytype(tangoTypeConst) *>(ptr_););
}
#else
template <long type>
static void _dev_var_x_array_deleter(PyObject *obj)
{
    void *ptr_ = PyCapsule_GetPointer(obj, NULL);
    TANGO_DO_ON_ATTRIBUTE_DATA_TYPE_ID(type, delete static_cast<TANGO_const2arraytype(tangoTypeConst) *>(ptr_););
}

template <>
inline void _dev_var_x_array_deleter<Tango::DEV_PIPE_BLOB>(PyObject *obj)
{
    // Unsupported
    assert(false);
}
#endif

template <long tangoTypeConst>
static inline void _update_array_values(Tango::DeviceAttribute &self, bool isImage, bopy::object py_value)
{
    typedef typename TANGO_const2type(tangoTypeConst) TangoScalarType;
    typedef typename TANGO_const2arraytype(tangoTypeConst) TangoArrayType;

    // Extract the actual data from Tango::DeviceAttribute (self)
    TangoArrayType *value_ptr = 0;
    try
    {
        self >> value_ptr;
    }
    catch(Tango::DevFailed &e)
    {
        if(strcmp(e.errors[0].reason.in(), "API_EmptyDeviceAttribute") != 0)
        {
            throw;
        }
    }

    static const int typenum = TANGO_const2numpy(tangoTypeConst);

    if(value_ptr == 0)
    {
        // Empty device attribute
        value_ptr = new TangoArrayType();
    }

    TangoScalarType *buffer = value_ptr->get_buffer();
    char *ch_ptr = reinterpret_cast<char *>(buffer);

    npy_intp dims[2];
    int nd = 1;
    size_t write_part_offset = 0;
    if(isImage)
    {
        nd = 2;
        dims[1] = self.get_dim_x();
        dims[0] = self.get_dim_y();
        write_part_offset = dims[1] * dims[0];
    }
    else
    {
        nd = 1;
        dims[0] = self.get_dim_x();
        write_part_offset = dims[0];
    }

    // Create a new numpy.ndarray() object. It uses ch_ptr as the data,
    // so no costy memory copies when handling big images.
    PyObject *array = PyArray_SimpleNewFromData(nd, dims, typenum, ch_ptr);
    if(!array)
    {
        delete value_ptr;
        bopy::throw_error_already_set();
    }

    // Create the numpy array for the write part. It will be stored in
    // another place.
    char *w_ch_ptr = 0;

    if(self.get_written_dim_x() != 0)
    {
        w_ch_ptr = reinterpret_cast<char *>(buffer + write_part_offset);
    }

    if(isImage)
    {
        nd = 2;
        dims[1] = self.get_written_dim_x();
        dims[0] = self.get_written_dim_y();
    }
    else
    {
        nd = 1;
        dims[0] = self.get_written_dim_x();
    }

    PyObject *warray = PyArray_SimpleNewFromData(nd, dims, typenum, w_ch_ptr);
    if(!warray)
    {
        Py_XDECREF(array);
        delete value_ptr;
        bopy::throw_error_already_set();
    }

    // numpy.ndarray() does not own it's memory, so we need to manage it.
    // We can assign a 'base' object that will be informed (decref'd) when
    // the last copy of numpy.ndarray() disappears.
    // PyCObject is intended for that kind of things. It's seen as a
    // black box object from python. We assign him a function to be called
    // when it is deleted -> the function deletes the data.
    PyObject *guard = PyCapsule_New(static_cast<void *>(value_ptr), NULL, _dev_var_x_array_deleter<tangoTypeConst>);
    if(!guard)
    {
        Py_XDECREF(array);
        Py_XDECREF(warray);
        delete value_ptr;
        bopy::throw_error_already_set();
    }
    PyArray_SetBaseObject(to_PyArrayObject(array), guard);
    py_value.attr(value_attr_name) = bopy::object(bopy::handle<>(array));

    // The original C api object storing the data is the same for the
    // read data and the write data. so, both array and warray share
    // the same 'base' (guard). Thus, the data will not be deleted until
    // neither is accessed anymore.
    if(warray)
    {
        Py_INCREF(guard);
        PyArray_SetBaseObject(to_PyArrayObject(warray), guard);
        py_value.attr(w_value_attr_name) = bopy::object(bopy::handle<>(warray));
    }
    else
    {
        py_value.attr(w_value_attr_name) = bopy::object();
    }
    // py_value.attr("__internal_data") = object(handle<>(borrowed(guard)));
}

template <>
inline void _update_array_values<Tango::DEV_STRING>(Tango::DeviceAttribute &self, bool isImage, bopy::object py_value)
{
    _update_array_values_as_tuples<Tango::DEV_STRING>(self, isImage, py_value);
}

template <>
inline void _update_array_values<Tango::DEV_ENCODED>(Tango::DeviceAttribute &self, bool isImage, bopy::object py_value)
{
    /// @todo Sure, it is not necessary?
    assert(false);
}

// template<long tangoTypeConst>
// static inline void _update_array_values(PythonDeviceAttribute &self, bool isImage)
// {
//     return _update_array_values_numpy<tangoTypeConst>(self, isImage);
// }

template <long tangoTypeConst>
static inline void
    _fill_numpy_attribute(Tango::DeviceAttribute &dev_attr, const bool isImage, const bopy::object &py_value)
{
    typedef typename TANGO_const2type(tangoTypeConst) TangoScalarType;
    typedef typename TANGO_const2arraytype(tangoTypeConst) TangoArrayType;

    PyObject *array = py_value.ptr();

    // -- Check dimensions
    Py_ssize_t dim_x = 0, dim_y = 0, nelems = 0;
    bool ok;
    switch(PyArray_NDIM(to_PyArrayObject(array)))
    {
    case 2: // -- Image
        ok = isImage;
        dim_x = PyArray_DIM(to_PyArrayObject(array), 1);
        dim_y = PyArray_DIM(to_PyArrayObject(array), 0);
        nelems = dim_x * dim_y;
        break;
    case 1: // -- Spectrum
        ok = !isImage;
        dim_x = PyArray_DIM(to_PyArrayObject(array), 0);
        dim_y = 0;
        nelems = dim_x;
        break;
    default: // -- WTF?!!?
        ok = false;
        break;
    }
    if(!ok)
    {
        raise_(PyExc_TypeError, isImage ? non_valid_image : non_valid_spectrum);
    }

    // -- Allocate memory for the new data object
    unique_pointer<TangoArrayType> value;
    CORBA::ULong unelems = static_cast<CORBA::ULong>(nelems);
    TangoScalarType *buffer = TangoArrayType::allocbuf(unelems);
    try
    {
        value.reset(new TangoArrayType(unelems, unelems, buffer, true));
    }
    catch(...)
    {
        TangoArrayType::freebuf(buffer);
        throw;
    }

    // -- Copy from numpy.array to TangoArrayType...
    PyArrayIterObject *iter;
    iter = (PyArrayIterObject *) PyArray_IterNew(array);
    if(!iter)
    {
        bopy::throw_error_already_set();
    }

    bopy::handle<> _h((PyObject *) iter);
    bopy::object iter_guard(_h);

    if(isImage)
    {
        // Why not use PyArray_ITER_NEXT() instead of PyArray_ITER_GOTO()?
        // We could do a single while(iter->index < iter->size) instead
        // of the double "for".
        // I did this and it worked in the sense that it went across
        // the correct number of elements but... I did not know the
        // x and y position it corresponded! Yes, 'iter' has a coordinates
        // field, but it was always [0,0], never updated!!
        npy_intp coordinates[2];
        npy_intp &x = coordinates[1];
        npy_intp &y = coordinates[0];
        npy_intp ndim_x = static_cast<npy_intp>(dim_x);
        npy_intp ndim_y = static_cast<npy_intp>(dim_y);
        for(y = 0; y < ndim_y; ++y)
        {
            for(x = 0; x < ndim_x; ++x)
            {
                PyArray_ITER_GOTO(iter, coordinates);

                PyObject *dataObj = PyArray_GETITEM(to_PyArrayObject(array), iter->dataptr);
                const bopy::object py_data = bopy::object(bopy::handle<>(dataObj));

                python_tangocpp<tangoTypeConst>::to_cpp(py_data, buffer[y * ndim_x + x]);
            }
        }
    }
    else
    {
        for(Py_ssize_t x = 0; x < dim_x; ++x)
        {
            PyObject *dataObj = PyArray_GETITEM(to_PyArrayObject(array), iter->dataptr);
            const bopy::object py_data = bopy::object(bopy::handle<>(dataObj));

            python_tangocpp<tangoTypeConst>::to_cpp(py_data, buffer[x]);

            PyArray_ITER_NEXT(iter);
        }
    }

    // -- Insert into device attribute
    dev_attr.insert(value.get(), static_cast<int>(dim_x), static_cast<int>(dim_y));

    // -- Final cleaning...
    value.release(); // Do not delete value, it is handled by dev_attr now!
}

template <>
inline void _fill_numpy_attribute<Tango::DEV_ENCODED>(Tango::DeviceAttribute &dev_attr,
                                                      const bool isImage,
                                                      const bopy::object &py_value)
{
    // Unsupported
    assert(false);
}
} // namespace PyDeviceAttribute