Crash when overloading returns custom class

Here is a minimum working example of my problem. It consists of two files in two directories, like this:

...
    myClass.m
    @double       (this is a directory)
        mtimes.m

I have defined a new class called myClass. Calling myClass(a, b), with a and b being doubles, returns a myClass object, with a 'num' and a 'str' propery. Both properties are just the product a*b, as a double and written as a string. Here is the code, saved in myClass.m:

    classdef myClass
        properties
            num
            str
        end

        methods
            function c = myClass(a, b)
                res = a * b;
                c.num = res;
                c.str = num2str(res);
            end
        end 
    end

I want to call myClass every time I try to multiply a double with a char. For this, I have the following code in the file called mtimes.m in a directory @double:

    function c = mtimes(a, b)
        if isa(a, 'double') && isa(b, 'char')
           c = myClass(a, str2double(b)); 
        else
            % Normal double*double multiplication:
            c = builtin('mtimes', a, b);
        end
    end

Now, myClass(3, 5) and 3*'5' are exactly equivalent (the same is true for any other pair of numbers). In the command window, everything works perfectly. I can run 3*'5' as well as x = 3*'5'. If I now create an m-file in the same directory as myClass.m and run 3*'5' from this file, it works the same. But when I assign it to a variable, x = 3*'5', Matlab crashes with a large error message.

Incidentally, if I do the same thing with a standard class instead of a custom written one, no error occurs.

What is the problem here? How can it be solved? And how come that the behavior of code inside an m-file is different from that of direct input into the command window?

If anyone can solve this, I would be very grateful!

Answer 1

[Matt J]

I'm just guessing, but it might be that the behavior is different inside an mfile because of the JIT. The syntax you use to generate x=3*'5' leaves no hints that the outcome will be other than double. The only way to determine the class of the result is to execute 3*'5' first. If the JIT tries to circumvent that mechanism in the interest of speed and predicts ahead of time that 3*'5' will yield the usual double result, then when you end up assigning to x a non-double data type, I can see it creating a segmentation fault.

[Matt J]

By the way, it appears that you've been way too persistent with this approach, ignoring many previous recommendations not to do this, e.g.,

http://www.mathworks.com/matlabcentral/answers/61673-variable-assignment-of-new-class-through-mtimes-works-in-command-window-but-not-in-m-files

http://www.mathworks.com/matlabcentral/answers/60410-overload-mtimes-for-double-char

Why is it worth all this pain? It seems unnecessarily complicated and dangerous to hack the native @double/mtimes method. As I advised you in one of those threads, why not create your own myChar class if you want new behavior for the char class? You're already using a user-defined class anyway.

[Jeppe]

Thanks for the response.

I have understood that what I'm trying to do is "dangerous". I'm quite sure though, that in the final implementation, it's safe. It is wort "all the pain" because I like the syntax 3*'5' much more than myClass(3, 5), in the real situation.

I have tried disabling JIT, but the result is the same. Also, if I set myClass to return a char or a uint or a cell or any other built-in data type, no problem occurs.

I can even use x = mtimes(3, '5') inside m-files. It is only the operator form x = 3*'5' which does not work.

I do not understand what you mean by creating my own myChar class. What should this class be like? How should I use it?

[Matt J]

I'm quite sure though, that in the final implementation, it's safe.

There's no real basis for your certainty. If you could be sure that there's a way to get it to work, you would be able to fix the segmentation faults yourself.

I do not understand what you mean by creating my own myChar class. What should this class be like? How should I use it?

Below is a classdef something along the lines of what Cedric advised you in a previous post, except that I allow you to define a unit to have weight 1 by default. Notice from my examples that once you've constructed once instance of a given measurement unit, you can use exactly the simple multiplicative syntax that you've been aiming for.

>> kg=myclass('kg'); %a unit of kilograms
>> x=3*kg

x = 

      3 kg

>> y=4*x

y = 

      12 kg

    classdef myclass

        properties
                num=1;
                units
        end

            methods
                function c = myclass(units, num)

                    c.units = units;
                    if nargin>1,
                     c.num=num;
                    end

                end

              function c=mtimes(num,obj_myclass)

                  c= obj_myclass;
                  c.num = num*c.num;

              end

              function display(obj) 

                  disp ' '
                  disp([inputname(1) ' = '])
                  disp ' '
                  disp(['    ' num2str(obj.num) ' ' obj.units]);
                  disp ' '

              end

            end 
        end

[Cedric]

"all the pain" is likely to be close to infinity in that case. Matt's suggestion is to use a custom class for one of the operands in the product, so you can catch the operation with a simple overload, e.g.

 x = 3 * myChar(5)

or

 x = 3 * myChar('28')

In these expressions, ntimes() can be overloaded in myChar so you don't have to mess up with double.ntimes().