Re: arrays of objects.

[Juan Pablo Carbajal]

On Tue, Feb 5, 2013 at 12:02 AM, ernst <address@hidden> wrote:
> Hi Juan,
>
> if you don't mind that the example I give you is not useful,
> i can give you a quite simple one:
>
> constructor:
>
> function num = pn(dNum)
>
>   switch (nargin)
>     case 0
>       dNum = 0.0;
>     case 1
>       % nothing to do
>     otherwise
>       print_usage();
>   endswitch
>
>   if ~isa(dNum, 'double')
>      error(strcat('invalid type: ',class(dNum)));
>    endif
>   num.dNum = dNum;
>   num = class(num, 'pn');
> endfunction
>
> As you can see, i just wrap a double in a class object.
>
> The converse, is also easy:
>
> function dNum = doubleValue(num)
>    dNum = num.dNum;
> endfunction
>
>
> With this setting, you can reproduce what I wrote before (see below),
> right?
>
> greetings, Ernst
>
>
>> On Mon, Feb 4, 2013 at 10:25 PM, ernst <address@hidden> wrote:
>>> Hi all,
>>> I want to make octave use another kind of number
>>> (just think of intervals, rationals, continued fractions.....) almost
>>> like using the builtin types.
>>> To this end, i have a class pn and constructor invocation like pn(3.4).
>>>
>>> This already works quite well, I can do some basic programming works in
>>> octave
>>> using my own numbers.
>>> I can even use vectors and matrices: [pn(2) pn3); pn(4) pn(5)] works.
>>>
>>> BUT: I have problems with higher dimensional objects.
>>> One example is arrayfun:
>>> On the one hand side,  a=arrayfun(@pn,rand(2,3,4)) seem to work,
>>> on the other hand, arrayfun(@doubleValue,a) does not work,
>>> raising
>>>
>>> "error: can't perform indexing operations for class type",
>>>
>>> although doubleValue(pn(3)) does the correct conversion to double as
>>> expected.
>>>
>>> I tried with arrayfun with doubles and with chars, both work well.
>>> Does it make sense, not to allow general classes?
>>> How does Matlab react?? (i dont have one, because it is too expensive).
>>>
>>> The alternative would be, to provide array-indexing for my class, but i
>>> think, this makes no sense,
>>> because i just want to use the indexing provided by octave.
>>> I have the idea it must be like for java or c, where array indexing
>>> is not needed to be implemented for each type.
>>>
>>> Help very much apprechiated.
>>>
>>> greetings,
>>>
>>> Ernst
>>> _______________________________________________
>>> Help-octave mailing list
>>> address@hidden
>>> https://mailman.cae.wisc.edu/listinfo/help-octave
>> Can't tell what is the issue without more information, but based on my
>> own experinece with cellfun, the old class style is not woring very
>> well inside those functions. Is as if one would need to declare these
>> functions "friend functions" as in C++.
>>
>> Though I can be completely off, not knowing exactly what is your
>> issue. Maybe you can upload your code to Agora?
>>
>

Hi ernst,

People around here like that answers are posted below the previous
message and not on top.

The problem you hit on is not the one I was referring to. As I
understand it, in your case, you should get an "invalid argument"
error rather than the one you get (more cryptic). Why? Since array fun
expects an array of type (from the help) logic, numeric, string,
structure or cell and you are passing an array of classes. For
example, this also would not work
 arrayfun(@length,[pn(1); pn(2)])
or any other execution of arrayfun with an array of classes.

It is ok that arrayfun is not working in your case, cause the use you
want is not implemented. Still there is the problem of the cryptic
error message.

If you have the time you could report the "Crpytic error message" when
passing an array of classes to arrayfun in the bug tracker, so this
issue is not lost. It is very basic to fix, so I will do it asap.

Cheers