Mesh.vertex_attribute doesn't work ass expected

Describe the bug I might be tired, or stupid… While looping through all vertices of a mesh and updating a vertex attribute of one vertex, all the vertices in the loop also get that value.

To Reproduce Run the following code:

from compas.geometry import Box
from compas.datastructures import Mesh

# make a simple compas mesh
mesh = Mesh.from_shape(Box.from_width_height_depth(2, 2, 2))

# add a new vertex attribute to all vertices
mesh.update_default_vertex_attributes({"new_attr": []})

# add a value to this new attribute only to some vertices
for index, v_attributes in enumerate(mesh.vertices(data=True)):
    print("------", index)
    v_key, v_data = v_attributes
    attribute = mesh.vertex_attribute(v_key, 'new_attr')
    print('before', attribute)
    if index % 2 == 0:
        attribute.append('why_{}'.format(index))
    print('after', mesh.vertex_attribute(v_key, 'new_attr'))

The output I get is:

------ 0
before []
after ['why_0']
------ 1
before ['why_0']
after ['why_0']
------ 2
before ['why_0']
after ['why_0', 'why_2']
------ 3
before ['why_0', 'why_2']
after ['why_0', 'why_2']
------ 4
before ['why_0', 'why_2']
after ['why_0', 'why_2', 'why_4']
------ 5
before ['why_0', 'why_2', 'why_4']
after ['why_0', 'why_2', 'why_4']
------ 6
before ['why_0', 'why_2', 'why_4']
after ['why_0', 'why_2', 'why_4', 'why_6']
------ 7
before ['why_0', 'why_2', 'why_4', 'why_6']
after ['why_0', 'why_2', 'why_4', 'why_6']

Expected behavior I’m expecting to change the attribute only of the desired vertex/vertices. An output like the following:

------ 0
before []
after ['why_0']
------ 1
before []
after []
------ 2
before []
after ['why_2']
------ 3
before []
after []
------ 4
before []
after ['why_4']
------ 5
before []
after []
------ 6
before []
after ['why_6']
------ 7
before []
after []

Desktop (please complete the following information):

• OS: macOS Catalina
• Python version: 3.7.6
• Python package manager: Conda
• Compas: 0.19.1

[gonzalocasas]

random idea: would it make sense to allow setting a function as default value, which acts as the constructor for complex types? Eg.

# as an anonymous function:
mesh.update_default_vertex_attributes({"new_attr": lambda: []})

# or as a named function
def create_default_box():
    return Box.from_width_height_depth(1, 1, 1)

mesh.update_default_vertex_attributes({"another_new_attr": create_default_box})

this would make the behavior requested in this issue possible without convoluting the client code (at the expense of a bit of extra complexity in the internals of attribute views, having to type-check and execute functions when a default is requested).

[gonzalocasas]

This is actually expected behavior from python. The empty list you pass as default is a reference. It’s the same assigned to all attributes effectively (the attributes are getting the memory reference to that list, not copies of the list).

Despite this being somewhat confusing, I would be tempted to leave as is because it’s also what happens when you use an empty list as the default value of a function argument.