Alternate Constructor Behavior with inputParser

Hi,

I have several classes whose constructors rely on InputParser to collect all the relevant properties. I feed all of the varargin array to the inputParser object. Then I loop over the results struct of the inputParser object and set each field of my class instance equal to the value of the matching field of the struct.

 parser.parse(varargin{:});
 results = parser.Results;
 field_list = fields(results);
 for field_index = 1:length(field_list)
   field = field_list{field_index};
   self.(field) = results.(field);
 end

While this works well and allows me to avoid complex syntax in the constructor (not to mention implementation logic), I don't have a good way to emulate multiple-constructor behavior. As I understand it, the standard way to have a non-default constructor is to switch on nargin inside the main constructor. With the inputParser approach, though, that's not reliable (as some inputs are optional, I can't predict nargin with any certainty). I could probably add an optional argument(s), but it's still awkward.

My typical use case is that I need to build an object directly from a string or input file with a syntax defined externally. While in other languages I'd probably use a secondary constructor that takes in either the file path or a string, in MATLAB the best I've come up with is to have a separate function called load_whatever.m that takes in the non-default constructor arguments and calls the default (only) constructor in the usual way. (This is slightly reminiscent of the Factory Design Pattern, but I don't know if the inventors of that one would approve.) Unfortunately, this involves moving class-specific code outside of the class definition file, and is neither object-oriented nor easily configurable.

The only other thing I can think of would be to have a static class method that could act as a secondary constructor. I have a feeling that's a bad idea on more than one count, so I haven't tried that yet.

What's the preferred practice for doing things in single-constructor languages like MATLAB that in multiple-constructor languages would use a second constructor? Is the thing I want to do here even one of those things?

Thanks, Jay

[Jiro Doke]

Can you elaborate on what type constructor signature you want to have? Is the example below your actual case, i.e. two input constructor and a structure input constructor?

I'm not promising anything, but maybe if we have an idea of what exact constructor behaviors you're looking for, we can try to figure it out.

Answer 1

[Jay]

So far, the static method is a leading candidate. Ruby has the same problem http://www.ruby-forum.com/topic/160572#706772

 %%classA.m
 classdef classA
   properties(Access = public)
     foo
     bar
   end
   methods(Access = public)
     function self = classA(f, b)
       self.foo = f;
       self.bar = b;
     end
   end
   methods(Access = public, Static)
     function result = alt_constructor(some_structure)
       result = classA(some_structure.foo_field,some_structure.bar_field);
     end
   end
 end
 %%test_classA.m
 a = classA(1, 0)
 b = classA.alt_constructor(struct('foo_field', 2, 'bar_field', 99))

>> test_classA

a =

classA

Properties:
  foo: 1
  bar: 0

Methods

b =

classA

Properties:
  foo: 2
  bar: 99

Methods

[Jiro Doke]

I'm kind of confused. Typically, if you want to allow an equivalent structure input, then I feel that the corresponding non-structure syntax should be a param-value pair. For example, one might use this syntax:

a = classA('foo', 1, 'bar', 0)

or this syntax:

b = classA(struct('foo', 2, 'bar', 0))

In this case, inputParser would take care of it with the "addParamValue" method.

[Jay]

Right, that example doesn't actually cover the first half of my rant. In most actual classes, I do call the primary constructor that way, or with a structure input. All this code does is demonstrate that it's possible for a static method to call a constructor. For some reason I was thinking that wouldn't work.