Struct member values

[charles-esterbrook]

Printing raylib structs gives opaque strings like <cdata 'struct Color' owning 4 bytes> whether using str() or repr().

The type is there, but not the member values. Whether labeled or not, they would be really useful.

Any chance for that?

[electronstudio]

You could probably format this a bit nicer but here is an example:

import pyray as pr

camera = pr.Camera3D([18.0, 16.0, 18.0], [0.0, 0.0, 0.0], [0.0, 1.0, 0.0], 45.0, 0)


def dump(struct):
    s = f"{pr.ffi.typeof(struct)}: ("
    for field in dir(struct):
        data = struct.__getattribute__(field)
        if str(data).startswith("<cdata"):
            data = dump(data)
        s = s + f"{field}:{data} "
    s = s + ")"
    return s


print(dump(camera))

[charles-esterbrook]

Nice. Here is a version which matches Python instantiation syntax more closely:

def structToStr(struct: object) -> str:
    # example output: 'Vector2(x=3.0, y=4.0)'
    name = pr.ffi.typeof(struct).cname.split(' ', 1)[-1]
    parts = [name, "("]
    sep = ''
    for field in dir(struct):
        data = struct.__getattribute__(field)
        dataStr = str(data)
        if dataStr.startswith("<cdata"):
            dataStr = structToStr(data)
        parts.append(sep)
        parts.append(f"{field}={dataStr}")
        sep = ', '
    parts.append(')')
    return ''.join(parts)

Any way this could become the default repr() of the various types?

If there are objects with too many fields, a keys argument could be used to select a subset. Just thinking out loud.

[electronstudio]

The cffi types are immutable and can't be monkey patched so I don't see an easy way of replacing repr(). Could perhaps create a proxy object and use that in place of the real object everywhere. But really this should be done by cffi, maybe ask them.

[charles-esterbrook]

I tried making a Vector2 wrapper class with the hooks to make it "tuple-like" via __len__, __getitem__ and __iter__, but when passed to raylib functions, they error with:

  File "", line 22, in render
    pr.draw_rectangle_v(self.loc, self.size, self.color)
  File "python3.12/site-packages/pyray/__init__.py", line 65, in func
    result = a(*modified_args)
             ^^^^^^^^^^^^^^^^^
TypeError: initializer for ctype 'struct Vector2' must be a list or tuple or dict or struct-cdata, not Vector2

Looks like specific types are required rather than "list-like" or "tuple-like". Is that something that can be relaxed?

I have other wrapper ideas in mind such as various methods and operator overloads.

[electronstudio]

Even if you make a wrapper that proxies every attribute it seems the cffi can still tell it's not a real cdata object. So then you would have to write code to detect your wrapper objects and unwrap them on every function call, and re-wrap every result returned back. It's possible, but the main motivation for the cffi binding was to keep things simple and let cffi do everything automatically. It's grown a little more complex over the years but wrapping/unwrapping everything would seem to be a lot of additional complexity. At some point it wouldn't be worth using cffi at all, you may as well just code everything manually like other bindings do. I could be wrong of course, there could be some easy way to do it I'm not aware of.

[charles-esterbrook]

If the cffi checked for hasattr on __len__, __getitem__ instead of isinstance(x, list) for accepting alternate argument types, then wrappers would work the same as lists ([3, 4]) and tuples ((3, 4)). I haven't yet looked into where the checks for the list and tuple are being specified and/or generated.

[electronstudio]

I’m not sure how much CFFI/ctypes are even Python objects. I think this stuff is implemented in C as part of CPython itself, so it’s not going to be easy to mess with from the Python side.

Also, something that has come up in the past, different Python implementations do CFFI/ctypes differently, so even if you get it working on CPython you might find it doesn’t work on PyPy.

[electronstudio]

We found a way of making a vector2 wrapper that seems to work in place of the real vector2: https://github.com/electronstudio/raylib-python-cffi/blob/master/examples/extra/vector2_extended.py