Why does strcmp take numerical arguments?

I found an old bug in my code. I pass by mistake a numerical value to strcmp. Now, I made a little test with R2013a

    num = double('A');
    is1 = strcmp( num, {'A','B','C'} );
    is2 = strcmp( num, {'A','B','C',num} );
    is3 = strcmp( num, {'A','B','C',num2str(num)} );
    [any(is1),any(is2),any(is3)]

returns

    ans =
         0     0     0

strcmp accepts the numerical argument, but doesn't find anything. I would have appreciated an error together with a message.

On the other hand, I use a similar behavior of strfind. It takes numerical inputs.

    n1  = (65:67);
    n2  = (61:70); 
    strfind( n2, n1 ) % us' trick

returns

    ans =
         5

That is an efficient way to search for sub-sequences in row vectors of flints (whole numbers). Last millennium the difference in speed mattered.

Answer 1

[Mike Hosea]

I think the intent was that the semantics of strcmp be

ischar(a) && ischar(b) && strcmp(a,b)

where in that expression it obviously doesn't matter what strcmp does if either a or b is not a string. That way, when processing inputs to a function you can just write

if strcmp(opt,'high road')

and be done with it. I know that this is pretty much what I have wanted strcmp to do, and been glad that it did, on a number of occasions. Anyway, once this was the behavior, it certainly became a backward compatibility issue to change it. It is inconsistent with strfind, unfortunately, but I think strfind's behavior is also a backward compatibility thing. The help text and doc don't mention supporting non-string input.

[Jan]

+1. Then the only problem is the incomplete documentation.

[Daniel Shub]

For r2013a the documentation for STRCMP, under tips, says "The strcmp function is intended for comparison of character data. When used to compare numeric data, it returns logical 0." Not where I would expect the warning to be, but it is there.

Answer 2

[Jan]

Another example:

strcmp(1, 1)    % FALSE
strcmp(pi, pi)  % FALSE

This behavior can be observed in 2011b, 2009a and version 6.5 also. Therefore it looks like a backward compatibility feature.