Arguments accept char, string or "cellstr" (cell array of char or string)

I have a large number of functions I eventually intend to be redistributable at my company. I had started creating my own variable validation routines but, fortunately I found the built in arguments block before I got too far down that road. It's very useful in data validation.

It's nice that arguments will allow not just strictly the given type/class of variable but will also allow anything that is convertible to that type (e.g. you can pass a string to char or char to string and any numeric types are also automatically converted). If you have two different types, it's a little trickier, but I've figured out how to use mustBeA and an array of strings of the types I want to accept.

One thing haven't figured out how to do effectively using this method. I have multiple aguments in multiple functions that can accept an input that is a char array, a string, or a cell array of char or string. I can use mustBeA to accept char, string, or cell - but I don't want to accept just any cell variable, it needs to be a cell array of char or string only, and it's not directly convertible. I know some MATLAB functions have sort of defined a type called cellstr which is a cell array of char or string - but that doesn't appear to be a checkable type in the arguments block.

I can, of course, add a block in my actual code to see if the argument matches iscell and then loop through the array and check all the elements but it would be nice if there's a way to do it within the arguments block.

Is there a way to do this inside the arguments block?

Answer 1

[dpb]

You write a custom validation function that your code in the arguments block calls for special cases if you're using the supplied arguments checking routines.

I've not really adopted it as don't write "production" MATLAB code for others to use, but my "trick" for the specific case you've outlined above looks like

function res=myfunction(arg)
  argOK=true;
  try
    cellstr(arg);       % throws error if any cell content isn't char or string
  catch
    argOK=false;
    ...                 % do whatever need here -- throw error, fixup if can, whatever...
  end
  ...

The less tricky, more straightforward way would be to use something like

  isOK=all(cellfun(@(c)ischar(c)|isstring(c),arg),'all');

in the custom arg test function.

[Captain Karnage]

Great answers.

Personally, I avoid using try/catch except in debug versions of my code so I would prefer not to use your first version.

Nice implementation of cellfun for a single line, also. While I was troubleshooting this, I found out I overlooked a validation function. mustBeText almost does this. Though your cellfun version will check for strings inside of cells, which mustBeText doesn't do.

So I may use that combined with @Les Beckham' s structure to make sure it returns what I need. My one concern is processor efficiency. I'm afraid if I'm calling cellfun in my validation function that gets called everytime my functions are called may end up slowing my programs down a lot. I'll code it up and see what works.

[Steven Lord]

Rather than using a cell array containing string arrays, I recommend just using string arrays.

cs = {'apple'; 'banana'}

cs = 2×1 cell array
    {'apple' }
    {'banana'}

s = string(cs)

s = 2×1 string array
    "apple"
    "banana"

c = char(s)

c = 2×6 char array
    'apple '
    'banana'

While you could define css as a cell array containing string arrays

css = cellfun(@string, cs, UniformOutput = false)

css = 2×1 cell array
    {["apple" ]}
    {["banana"]}

Your code would need to "peel" the banana out of css using curly braces if it wanted to do much with it. In my experience, indexing with parentheses tends to be easier for users to understand than indexing with curly braces.

strlength(css{2})

ans = 6

strlength(s(2))

ans = 6

[Captain Karnage]

Don't get me started on curly braces. I'm working on converting some other code from cell arrays to standard arrays. Also tends to execute more efficiently. I'm finding out that you really only need cells if you have heterogeneous data (or - one exception - if you want a column of char vectors of different lengths), and even then sometimes you can avoid that with matlab.mixin. I've also found that using table and timetable without being careful will automatically create cells sometimes when it doesn't necessarily have to and trying to avoid that as well.

However, I want to support this type being passed to my function as I'm doing this from the perpsective of "what could other users feed to my function"

[Matt Cooper]

The cellfun method should also check for cell arrays of cells containing text:

tf = all(cellfun(@(x)ischar(x)|isstring(x)|iscellstr(x),x),'all');

Answer 2

[Les Beckham]

Interesting issue. Looks like you have to create a "custom validation function". See if something like this works for you.

a = ones(1, 3);
Test(a)

Error using solution>Test
Invalid argument at position 1. inputArg1 must be string or char or cellstr

Test('a')
Test("a")
Test({'a', 'b'})
Test({"a", "b"})

function Test(inputArg1)
arguments
    inputArg1 {mustBeStringCharOrCellstr(inputArg1)} 
end
disp(inputArg1)
end

function mustBeStringCharOrCellstr(arg)
    if ~isstring(arg) && ~ischar(arg) && ~iscellstr(arg)
        eid = 'Test:notValid';
        msg = 'inputArg1 must be string or char or cellstr';
        throwAsCaller(MException(eid, msg))
    end
end

[Captain Karnage]

You know it's funny... iscellstr actually just specifically checks for a cell array of char. If it's a cell array of string it actually returns false. It's one of a handful of what I would call "badly named MATLAB functions". I'm sure it has to do with the history of MATLAB and not officially having a string type for a long part of it's history and essentially having cell arrays of char were the only way to represent a string and were casually referred to as strings.

[Captain Karnage]

So... I love your answer... but I was about to adapt this code and while looking up something else, I found an overlooked already existing function! mustBeText does the same thing... (also not including an actual cell array of strings, but I think I can live with that as I typically don't use that form factor and I'll probably just remove support for it from my functions)

[Les Beckham]

Good to know. I actually haven't used the argument validation feature very much, so it is nice to learn new things about it so that, if and when I do use it, I will be more prepared.

Creating a cell array of strings, to me, sounds like just poor design, since you can already have arrays of arbitrary length strings. Note that the Matlab editor will even flag the creation of a cell array of strings with a Code Analyzer warning, and offer to fix it by converting it to an array of strings (square brackets instead of curly ones.)

So, maybe you should answer your own question with the mustBeText solution and accept that answer so people looking at this in the future will know about that solution. Note that you can also vote for answers that helped you (such as dpb's and mine) even if they aren't the "right" or accepted answer.

Answer 3

[Captain Karnage]

If cell arrays of strings wasn't a requirement, MATLAB already has an easy built-in solution with mustBeText:

arguments
    thisarg     { mustBeText(thisarg) };
end

To support cell arrays of strings - I've bobbled back and forth on what method I wanted to use. In the end I created a custom validation function derived from the source code of MATLABs mustBeText function. I added acceptance of a cell array of strings andI took it one step further and made it validate with any number of layers of cells. It has a safe recursion (by "safe" I mean it takes a layer off each time and so eventually must reach an end and cannot create an infinite loop). I struggled with naming this dang thing in a non-overbearing way that also says what it does, but here is my end result:

function mustContainOnlyText(arg)
%mustContainOnlyText validates that input contains only text
%   argument is char, string, or a cell array of strings
%   Intended for use within arguments block to validate an input

    if ~containsOnlyText(arg)
        eid = 'custom:validators:mustContainOnlyText';
        msg = 'Value must be char, character vector, string array or scalar, or cell array of character vectors or strings.';
        throwAsCaller(MException(eid, msg));
    end

end

which calls the function:

function tf = containsOnlyText(text)
    %containsOnlyText - checks that passed text contains only text
    %   In this context:
    %   "text" refers to char or string types
    %   "contains" means it could be text by itself, 
    %       or either type contained within an array or cell
    %   Returns true if it contains only text
    
    %Single Boolean Statement
    %Removes outer cell layer and recurses if the passed text is in a cell array
    %Text passed to next version is horizontally concatenated version
    tf =    ischar(text) || ...
            isstring(text) || ...
            iscell(text) && ( containsOnlyText( [text{:}] ) ); %If there are layered cells
end

and I can then add to the arguments block:

arguments

    thisarg    { mustContainOnlyText(thisarg) };

end

I'll probably tweak this some - I know that if someone passes a heterogenous array of cells it will cause an exception in the containsOnlyText function and not send the validation exception message (but, ultimately, it will still check for the correct type).

I'm contemplating adding try / catch in per @dpb. I'm normally opposed to a Try/Catch in "production" code because it could mask an exception - and I believe you should work to make the exception not possible instead of allowing the exception then patching around it. In this case though, try / catch would only serve to change the exception to a different exception.

Thank you to both @dpb and @Les Beckham who helped get me to this answer. And also yes, simple arrays of strings are much better than cell arrays of strings.

[Les Beckham]

You are welcome.

Are you sure that this code works? Your so called recursion looks like it is calling a function (isAnyText) that you didn't post. Or, perhaps, containsOnlyText should be calling itself instead?

[Captain Karnage]

I changed the name of the function several times and forgot to update the text in the code in this post. D'Oh. I just edited it to fix that.

[Les Beckham]

I suspected as much, but had to ask :).