Empty array of class objects from string name of class

If I have a class called SomeClass, I can create an empty array of SomeClass using:

arr = SomeClass.empty

Is there a way to create an empty array if I only have the character string "SomeClass".

[SK]

I suppose

eval("SomeClass.empty")

should do the trick.

[Image Analyst]

I think you mean

str = "SomeClass";  % Here is the character string.
eval(sprintf('arr = %s.empty', str)); % Put into command string and pass to eval()

[SK]

Yes. I was too lazy to put "SomeClass" into a variable.

[Stephen23]

Although slightly different from the usual approach, this will have almost all of the same disadvantages as the usual dynamic variable names that some beginners like to use to force themselves into writing slow, complex, buggy code:

https://www.mathworks.com/matlabcentral/answers/304528-tutorial-why-variables-should-not-be-named-dynamically-eval

[SK]

This is very true. Often, one can find a cleaner solution without dynamic types, variable names and so on.

Thank you for your comment. I had a slightly tricky problem involving construction of nested objects, and I realized that it would be easier to let the user call the constructor using a list of empty objects rather than a list of strings.

Update: FAIL! See my answer below.

Answer 1

[Guillaume]

"Those familiar with Matlab will know that this will not work"

This.elements(end+1) = obj;

Indeed, but the semantically equivalent

This.element = [This.element, obj];

will. No need to test for emptiness.

[SK]

Aah, I didn't know that (or at least didn't think of it, given that I've probably used that "feature" implicitly many times) - resulting in another time-consuming circular walk in woods. Thanks.

Answer 2

[Image Analyst]

Doing this - making variables named from strings - is a bad idea. I mean, just how would you refer to that variable later? How would you use it if you don't know the variable name before run-time? See this link

[Steven Lord]

It's extremely easy to come up with a piece of text that is not a valid class name but if used with Image Analyst's suggestion would do various nasty things. I'm not going to post one because I don't want people to get angry if they closed their MATLAB session by running the example.

If you must use this approach to run code from your users, make absolutely sure to sanitize your inputs. Use meta.class.fromName to validate that the class name you've been given is an actual class. Here's one that does not exist:

name = 'thisClassDoesNotExist';
if isempty(meta.class.fromName(name))
  fprintf('The class %s does not exist!\n', name);
else
  fprintf('Class name %s exists, you may create an empty.\n', name);
end

Here's one that does.

name = 'table';
if isempty(meta.class.fromName(name))
  fprintf('The class %s does not exist!\n', name);
else
  fprintf('Class name %s exists, you may create an empty.\n', name);
end

[SK]

Thank you for your replies. Please see my comment to Stephen Cobeldick https://in.mathworks.com/matlabcentral/answers/415310-empty-array-of-class-objects-from-string-name-of-class#comment_601103

Answer 3

[SK]

OK, so it turns out, that I cannot do what I wanted to accomplish. The original problem arose from my not wanting to write an extra (ugly) four lines of code.

Suppose I have a container class:

   class Container
     properties
       elements;
     end

     methods
       function  This = Container()
       end

       function This = push(This, obj)
         This.elements(end+1) = obj;
       end
     end
   end

Those familiar with Matlab will know that this will not work, as Matlab initializes "elements" to an empty "double" type and I cannot push another object into that array. Of course the problem is easily solved by rewriting push as:

       This = push(This, obj)
         if isempty(This.elements)
           This.elements = obj;
         else
           This.elements(end+1) = obj;
         end
       end

But this grated on the senses, so I thought, why not indicate the type of "elements", in the constructor, like:

       function  This = Container(typename_string)
         This.elements = eval([typename_string, '.empty']);
       end

or:

       function  This = Container(empty_array_of_desired_type)
         This.elements = empty_array_of_desired_type;
       end

Which is a nice solution allowing me to write the push method cleanly. The problem arises when "element" objects can also be containers (of a different type in the same hierarchy) and the elements of "elements" can also be containers - in other words there could be a certain amount of nesting. There is no way to provide all the right empty types to all levels without using eval() to dynamically create empty arrays of the right type, and even then it gets ugly with all the checking.

Conclusion: Bite the bullet and tolerate a little ugliness in the push method.