I can't test this on Windows

@MaximSmolskiy I have tested it under Windows using MSVC and there should be no memory leak anymore.

But I just did a very basic test, I executed the imread function 100 times to read the jxl file, and the memory usage gradually went up to 659M before this PR, and after this PR, the memory usage was only 5M.

i confirm the test pass on windows

frankly i did not understand what is the difference calling release() or reset()

    pointer release() noexcept {
        return _STD exchange(_Mypair._Myval2, nullptr);
    }
    void reset(pointer _Ptr = nullptr) noexcept {
        pointer _Old = _STD exchange(_Mypair._Myval2, _Ptr);
        if (_Old) {
            _Mypair._Get_first()(_Old);
        }
    }

frankly i did not understand what is the difference calling release() or reset()

As I understand it, reset() releases the original resource managed by unique_ptr and initializes unique_ptr with a new pointer (nullptr by default). release() only releases ownership of the resource, returns the raw pointer to the caller, and then sets unique_ptr to nullptr.

When release() is called without saving its return value, the sole reference to the internal resource (the raw pointer) is lost, making the memory inaccessible and resulting in a memory leak.

I think I understand, probably
_Old is FILE * and _Mypair._Get_first() is fclose so the file closed.

        if (_Old) {
            _Mypair._Get_first()(_Old);
        }

Yes. The first element of _Mypair is a deleter and the second element is raw pointer.

When JpegXLDecoder is initialized, the deleter for m_f is set to fclose, so it can then be deleted correctly with reset().