Fix default vdims in raster.RGB (round two)

[hoxbro]

Follow up to #5773

This is how RGB worked before:

import param

class Dim: pass

class RGB(param.Parameterized):
    value = param.List(default=[Dim(), Dim()])

    def __init__(self, value=None, **params):
        if value is None:
            value = list(self.value)

        super().__init__(value=value, **params)

So if Example() was run, value would use a copy of the self.value with the default values in it. So running something like this would only change the id of the list, but not of the values inside of the list:

By changing value = list(self.value) to value = copy.deepcopy(self.value), we get the expected behavior, a unique ID for everything.

And if we set value = self.value, we also get the same identity for the list, which is something we really do not want.

Because we are setting the value directly before calling super(), we are bypassing Param's mechanism for deep copying the default values.

[codecov-commenter]

Codecov Report

Merging #5775 (2cd76d9) into main (be58b7e) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##             main    #5775   +/-   ##
=======================================
  Coverage   88.16%   88.17%           
=======================================
  Files         307      307           
  Lines       62839    62850   +11     
=======================================
+ Hits        55405    55416   +11     
  Misses       7434     7434           

Flag	Coverage Δ
ui-tests	22.41% <38.46%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
holoviews/element/raster.py	78.97% <100.00%> (+0.04%)	⬆️
holoviews/tests/element/test_raster.py	100.00% <100.00%> (ø)

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[maximlt]

Thanks for the quick fix, it looks good to me.

I find this Param behavior pretty error prone and I wouldn't be surprised to find other similar bugs through the code base.

The __init__ below looks like a no-op but effectively overrides the deepcopied list (yep it's deepcopied anyway which isn't super efficient):

import param

class Foo(): pass

L = [Foo(), Foo(), Foo()]

class P(param.Parameterized):
    l = param.List(default=L)

    def __init__(self, l=None, **params):
        if l is None:
            l = self.l
        super().__init__(l=l, **params)

p = P()

assert P.l is not p.l  # Fails :(

Another way to fix that kind of thing is to avoid passing the list up to the super class if it's not modified, not to confuse Param's machinery, with something like that:

import param

class Foo(): pass

L = [Foo(), Foo(), Foo()]

class P(param.Parameterized):
    l = param.List(default=L)

    def __init__(self, l=None, **params):
        if l is None:
            # l = list(self.l)
            l = self.l
        else:
            l = list(l) if isinstance(l, list) else [l]
        if l is not self.l:
            params['l'] = l
        super().__init__(**params)

p = P()

assert P.l is not p.l  # Passes
assert all(a is not b for a, b in zip(P.l, p.l))  # Passes

I don't think there is much we can do at the Param level, besides documenting this behavior (it could avoid the deepcopy, maybe).