Here is a simple modification of the existing run.m that behaves properly on the example of comment #8:

function run (script)

  if (nargin != 1)
    print_usage ();
  endif

  [d, f, ext] = fileparts (script);

  if (isempty (d))
    % Do not change directory if the input argument has no path
    evalin ("caller", f, "rethrow (lasterror ())");
  else
    if (exist (d, "dir"))
      startdir = pwd ();
      d = make_absolute_filename (d);
      unwind_protect
        cd (d);
        evalin ("caller", f, "rethrow (lasterror ())");
      unwind_protect_cleanup
        if (strcmp (d, pwd ()))
          cd (startdir);
        endif
      end_unwind_protect
    else
      error ("run: the path %s doesn't exist", d);
    endif
  endif
endfunction

What do you think?

I wrote below that "Matlab changes directory in all cases", and this indeed what the documentation of Matlab R2016a says (currently online at: http://fr.mathworks.com/help/matlab/ref/run.html).

Unfortunately, it seems that the truth is more complicated (arghhhhh). Here is the output of the attached example script in Matlab R2016a:

=== Calling mfunc directly ===
myfunc has been called with 0 argins
current directory is: /home/bect/code/octave/bugs/47807/essai1
myfunc has been called with 2 argins
current directory is: /home/bect/code/octave/bugs/47807/essai1
=== Calling mfunc with run ===
myfunc has been called with 0 argins
current directory is: /home/bect/code/octave/bugs/47807/essai1/subdir
myfunc has been called with 0 argins
current directory is: /home/bect/code/octave/bugs/47807/essai1/subdir
myfunc has been called with 0 argins
current directory is: /home/bect/code/octave/bugs/47807/essai1
myfunc has been called with 0 argins
current directory is: /home/bect/code/octave/bugs/47807/essai1

It seems that Matlab actually changes directory in all cases but one: it doesn't change directory when the input argument has no "path part" (the first output arg of fileparts is empty).

This is easy to implement,

So should run be just as brain-dead simple as

function run (file_name)
curr_dir = pwd ();
onCleanup (@() chdir (currdir));
## Some code to check validity of file_name and determine directory part (if any) and function name...
chdir (dir_part_of_given_file_name);
eval (function_name_derived_from_file_name);
endfunction

?  Does the evaluation of the function need to happen in the parent (so instead of eval, use evalin ("caller", ...))?

It occurs to me that the more complicated case might be if Matlab allows this to work on a function file that is not in the path. So I should amend saying it looks easy to support the cause of "run FUNCNAME", but not to support "run /path/to/some/funcfile.m". Does Matlab support that (if the directory is not in the load path)?

I think this has been known and ignored for a while, but it seems simple enough to add if it's an important feature to someone.

Does Matlab also change directory to the location of the function file when it is executed? What's the output for a function /some/path/foo.m

function foo()
disp(pwd);
end

Does it show an error for functions that do require arguments? Do errors occur in the same context as if the function were just called without run? Can errors be caught with try or shown with lasterr?

The run() function of Octave treats "function m-files" (i.e., m-files where all the code appears in function/end[function] pairs) differently form its Matlab couterpart:

• Matlab executes the function without any input argument,

• whereas Octave executes the m-file as a script that defines one or several functions.