setdiff error in some cases

Hi I am using setdiff in my code and i am getting 'Improper assignment with rectangular empty matrix' error in a particular case , why and how to solve this? my code is as below:

 A = dataset();
 A.Location = {'south'};
 A.Number = 3;

 B= dataset();
 B.Location = {'north'};
 B.Number = 3;

 diff1 = setdiff(A, B); %THIS EXECUTES JUST FINE

 C= dataset();
 C.Location = {'north'};
 C.Number = 3;
 C(1, :) = [];

 D= dataset();
 D.Location = {'west'};
 D.Number = 5;

 D.Location(2, 1) = {'north'};

 diff2 = setdiff(C, D); %THIS EXECUTES JUST FINE

 diff3 = setdiff(C, B); %THIS GIVES ME ERROR MENTIONED ABOVE, WHY?

[Stephen23]

This would be a great opportunity to use MATLAB's inbuilt debugging tools Set a breakpoint on/before the line that causes the error, and use the debugger workspace to investigate the variables (sizes, values, etc).

[per isakson]

What is the intention of

    C(1, :) = [];

[dpb]

I'm not terribly surprised setdiff fails with an empty set as one of the inputs; the specific failure seems to be related to there being more than one element in the non-empty set given the case that succeeds.

Looks to me like one should special-case the code using the operation if it's necessary to have the case.

Always in a case such as this, ultimate response is TMW's interpretation of what their internal definition of accepted behavior should be so submit a support request at www.mathworks.com since it isn't documented.

[Neesha]

Stephen Cobeldick : Yes I have, and that is how i came up with example here with the code.

per isakson : just trying to delete a row, to end up with one row in the dataset

dpb : I am not sure what you meant by 'element' but i think you are describing the problem correctly that setdiff fails when one is empty and not-empty has one row. it does work though with empty dataset if non-empty ha more than one row. What is TMW?

[dpb]

...What is TMW?...

The Mathworks, vendors of Matlab. Shorthand for the suggestion you contact official support at www.mathworks.com with a support request for an official interpretation of the behavior observed if you think it's important that the null input case not fail.

Answer 1

[Peter Perkins]

Neesha, this appears to be a bug when one of the datasets has zero rows and the other has exactly one. A valid edge case to be sure, but somewhat obscure. I will make a note to adress this bug.

[Neesha]

Thanks, i have contacted technical support and submitted this.

ref:_00Di0Ha1u._500i0DabIS:ref