vertcat of values stored in structure

I prepared this simple classdef.

classdef myclassA

      properties
          prop
      end

      methods

          function obj = myclassA(val)
              obj.prop = val;
          end

          function out = horzcat(varargin)
              disp('horzcat was invoked');

              out = builtin('horzcat',varargin{:});
          end

          function out = vertcat(varargin)
              disp('vertcat was invoked');

              out = builtin('vertcat',varargin{:});

          end
      end
  end

Then I performed concatenation of objects.

>> obj1 = myclassA(1);
>> obj2 = myclassA(2);
>> [obj1,obj2]

horzcat was invoked

ans = 

1x2 myclassA array with properties:

    prop

>> [obj1;obj2]
vertcat was invoked

ans = 

2x1 myclassA array with properties:

    prop

So far so good. However, when I put the same objects into a structure, I've got rather unexpected behaviours of vertcat.

>> S.obj1 = obj1;
>> S.pbj2 = obj2;
>> [S.obj1,S.obj2]
horzcat was invoked

ans = 

1x2 myclassA array with properties:

    prop  

>> [S.obj1;S.obj2]
horzcat was invoked
horzcat was invoked
vertcat was invoked

ans = 

2x1 myclassA array with properties:

    prop

Here, the two objects were processed by horzcat twice and then by vertcat once. I wonder if this is expected behaviour (I doubt it) or it is a bug.

When I used vertcat explicitly rather than [a;b], I obtained results in a expected way.

>> vertcat(S.obj1,S.obj2)]
vertcat was invoked

ans = 

2x1 myclassA array with properties:

    prop

MATLAB R2015b, on Mac OS X El Capitan

Answer 1

[Walter Roberson]

When you use S.obj1 then it needs to do structure expansion [S(1).obj1, S(2).obj1, S(3).obj1, ... S(end).obj1)] and that is where the horzcat comes from. If you had used S(1).obj1 then horzcat would not have been invoked because there would be no structure expansion.

When there is only one object in the structure it does seem to be a waste to go through structure expansion. I do not know if there is some subtle reason that it must do so, but that is what appears to be happening.

[Kouichi Nakamura]

Oh, thanks a lot. This explains everything. MATLAB assumes S may be non-scalar structure! I thought obj1 and S.obj1 were equivalent and felt that behaviour was weird.