Recursive nested function gives different result on first run

I have a recursive function that is nested in my main function and when assigning its output to some variable it gives a different result (for certain inputs). Here's an example showing the behaviour

function arr = foo( )
%    clear all % makes it give the same (possibly wrong) result every time
   function arrB = foorecurse( currIds )
        disp(currIds)             % correct
        disp(cellfoo{currIds(1)}) % correct
        fprintf('---------\n')
        if length(currIds) == 1
            arrB = cellfoo{currIds}
        else
            arrB = [cellfoo{currIds(1)} ; foorecurse(currIds(2:end))]
        end
    end
    cellfoo = { eye(2), [0 1 ; 1 0], [0 -i ; i 0], [1 0 ; 0 -1] };

%     foorecurse([2 3 2])       % not assigning the output to any variable "fixes" it
    arr = foorecurse([2 3 2])   % gives a column of cellfoo{[3 3 2]} first time it's run
%     arr = foorecurse([2 1 2]) % this is fine
end

I have not tested it on another machine. I am at a total loss as to what could be happening. I know there's a way to achieve the same thing in many other ways (and they work) but for the sake of understanding what ML does, can anyone figure this out?

[Matt J]

Bizarre!

Answer 1

[Matt J]

I think it's a parsing bug and I encourage you to report it to Tech Support.

Initially, I supposed that it's some hazard of mixing recursive calls with concatentation operations, but this variation seems to work just fine,

   function arrB = foorecurse( currIds )

        if length(currIds)==1
            arrB = cellfoo{currIds};
        else

            arrB = [foorecurse(currIds(1)) ; foorecurse(currIds(2:end))];
        end
   end

[Aleksandrina Nikolova]

Well spotted, makes it even more bizarre! I reported it to Tech Support, will post any updates.

[Aleksandrina Nikolova]

You were right, it is a bug. I heard from the Technical Support:

"[T]he error that you face is a bug due to JIT implementation. However, it does not exist from the R2015b release of MATLAB. [...] [Y]ou could add the following line [to] turn off the JIT for the duration of the offending function."

feature scopedaccelenablement off

[Matt J]

Good to know!