Memory Leak using ExtractImageChips<T>(Array2DBase, IEnumerable<ChipDetails>) about dlibdotnet HOT 5 CLOSED

That's right.
For now, array class will not be released.

DLLEXPORT void array_delete(void* obj)
{
    // dlib::array is template type.
    // Ex. dlib::array<dlib::matrix<float, 31, 1>> does NOT equal to dlib::array<dlib::matrix<float>>
    // Template argument is decided in compile time rather than runtime.
    // So we can not specify template argument as variable.
    // In other words, we have to call delete for void* because we do not known exact type.
    // Unfortunately, dlib::array implement destructor.
    // What should we do?
    delete obj;
}

However, I probably should check type and delete it for except for template types looks like dlib::array<dlib::matrix<float, 31, 1>>.

from dlibdotnet.

I fixed it for me by modifying the wrapper:

` protected override void DisposeUnmanaged()
{
base.DisposeUnmanaged();

        switch (_ItemType)
        {
            case ItemTypes.PixelType:
                {
                    Dlib.Native.array_delete(this.NativePtr);
                }
                break;
            case ItemTypes.Array2D:
                {
                    var type = this._ArrayElementType.ToNativeArray2DType();
                    Dlib.Native.array_array2d_delete(type, this.NativePtr);
                }
                break;
            case ItemTypes.Matrix:
                {
                    throw new NotSupportedException("Memory leak");
                }
                break;
            default:
                throw new ArgumentOutOfRangeException();
        }
    }`

and adding the following native method:

DLLEXPORT void array_array2d_delete(array2d_type type, void* obj) { switch (type) { case array2d_type::UInt8: delete ((dlib::array<array2d<uint8_t>>*)obj); break; case array2d_type::UInt16: delete ((dlib::array<array2d<uint16_t>>*)obj); break; case array2d_type::Int16: delete ((dlib::array<array2d<int16_t>>*)obj); break; case array2d_type::Int32: delete ((dlib::array<array2d<int32_t>>*)obj); break; case array2d_type::Float: delete ((dlib::array<array2d<float>>*)obj); break; case array2d_type::Double: delete ((dlib::array<array2d<double>>*)obj); break; case array2d_type::RgbPixel: delete ((dlib::array<array2d<rgb_pixel>>*)obj); break; case array2d_type::HsiPixel: delete ((dlib::array<array2d<hsi_pixel>>*)obj); break; case array2d_type::RgbAlphaPixel: delete ((dlib::array<array2d<rgb_alpha_pixel>>*)obj); break; } }
I also did that for the matrix class, where you already had this code but as comment because of some special fixed size matrix templates. I had no issueswith it and I think at least the above array code is save.
And it fixed my memory leak. In case you want to integrate that.

from dlibdotnet.

Thank you so much!!
Please give me moment to check your codes.
Unfortunately, I am busy here recently, so I can not do any coding :(

from dlibdotnet.

OK, It should be ok.
Could you send PR if you can?
After merge, I add code to release array.

from dlibdotnet.

Fix by 5c34efa

from dlibdotnet.