Can I use Matlab's default implementation of empty with matlab.mix​in.indexin​g.Redefine​sParen?

I'm trying to use the new method of overloading indexing with the matlab.mixin.indexing.RedefinesParen class, new in 2021b. However, I like and am fine with Matlab's default implementation of empty, which becomes Abstract and requires definition in classes that inherit from RedefinesParen. In fact, I'm making a general top-level class and I'm trying to make things easy on developers, so I like that the default implementation of empty for classes doesn't have to worry about property validation, which is sometimes defined in the constructor rather than with the property definition.

Overloading empty and choosing an "empty" property that will pass through error-checking in the constructor is an extra layer of worry that I am hoping developers will not have to worry about when inheriting from this class. Is there any way to use the default empty that Matlab implements behind-the-scenes for classes that inherit from RedefinesParen or am I out-of-luck on this front?

Answer 1

[James Lebak]

You do have to provide an 'empty' for your class that inherits from RedefinesParen. Such classes are scalars masquerading as arrays, and MATLAB unfortunately can't guarantee that its default empty will be functionally correct in all cases for such classes.

If you provide your own implementation of 'empty', then classes that inherit from your class should not need to do anything extra.

[Robin]

So what would such an implementation look like if I just want the default Matlab behaviour? I cannot simply do

function obj = empty(varargin)

obj = MyClass.empty();

end

to create an empty object, because that results in an infinite recursion.

Also, varargin does not contain any information on the subclass that it should construct an empty array of.

[James Lebak]

Empty takes size arguments, either a size vector or a set of scalars:

https://www.mathworks.com/help/matlab/ref/empty.html

As specified in the documentation, at least one size parameter is expected to be zero when calling the empty method.

Classes that inherit from RedefinesParen are scalar objects masquerading as arrays. As such the usual implementation is that there are one or more properties in the object that store array data. The idea of the empty method is to construct those properties with the sizes passed to the empty method. We provide a couple of examples:

https://www.mathworks.com/help/matlab/matlab_oop/customize-parentheses-indexing-for-mapping-class.html

https://www.mathworks.com/help/matlab/ref/matlab.mixin.indexing.redefinesparen-class.html

It is true that varargin does not give you any information on subclasses in this context. If you need such information to correctly construct an empty of a given subclass then empty does have to be defined by the subclass.

[Robin]

Thanks James for the elaboration!

Unfortunately, my class is not a scalar object masquerading as an array and I have dozens of subclasses for the class that I am defining the empty method for, so defining it in each of those is not an option. I will see if using good ol' subsasgn is going to work for me.