Calling a nested function from a dynamic regular expression throws an error

Calling a local function from a dynamic regular expression works without error:

function test()
regexp('123456','\d(?@subfun($&))')
end
function subfun(s) %#ok<DEFNU>
    disp(s)
end

and prints the expected sequence in the command window:

>> test
1
2
3
4
5
6
ans =
     1     2     3     4     5     6

But calling an identical nested function:

function test()
   function subfun(s) %#ok<DEFNU>
       disp(s)
   end
regexp('123456','\d(?@subfun($&))');
end

throws this error:

Error using regexp
Attempt to add "matched" to a static workspace.
 See MATLAB Programming, Restrictions on Assigning to Variables for details.
Error in test (line 5)
regexp('123456','\d(?@subfun($&))');

Some experimentation showed that "matched" refers to the '&$' matched string token. I hypothesize that the "static workspace" referred to is somehow internal to regexp.

The error message's hyperlink goes to a page "Variables in Nested and Anonymous Functions", which states "If you attempt to dynamically add a variable to the workspace of an anonymous function, a nested function, or a function that contains a nested function, then MATLAB® issues an error of the form Attempt to add variable to a static workspace."

Given that there are no allocations within the nested function, why does this error occur? How is it possible to call a nested function (with dynamic input arguments) from a dynamic regular expression? I am using MATLAB R2012b.

Answer 1

[Guillaume]

Hum, interesting problem! The documentation of dynamic regular expressions is unfortunately lacking on the workspace used. As far as I can tell it's not a workspace private to regexp but the actual workspace of the function it executes in.

For example,

regexp('', '(?@ y = 5)')

will create y in the local workspace. So the problem is that regexp is trying to create the matched variables in the workspace of the function and as the documentation says, you can't create dynamic variables in functions with nested functions.

Don't know why it try to create that variables. I guess it's an implementation detail, where at some point &$ is replaced by a valid variable name before the expression is evaluated. Note that if you actually name a variable matched, then matlab tries to use matched_0 instead. It kind of makes sense, but yes, completely undocumented. I guess it's worthy of a bug report to mathworks.

You get the same problem in anonymous functions for the same reason:

>>fn = @() regexp('1234', '\d(?@ disp($&); disp(matched))')
>>fn
Error using regexp
Attempt to add "matched_0" to a static workspace.
 See Variables in Nested and Anonymous Functions.

It's still the same behaviour in 2015b and newer.

For info, still a problem in R2018a. Also: fixed a typo in my example!

[Stephen23]

Hmmmm... so two things happen:

1. dynamic regular expressions create some temporary variables in the workspace.
2. nested functions mean workspaces cannot have variables added dynamically...

This implies that the dynamic regular expression does not even need to call the nested function, it is enough for that nested function to simply exist. And yes, this turns out to be the case, confirming your hypothesis:

function test()
   function subfun(s)
       disp(s)
   end
regexp('123456','\d(?@$&)');
end

causes the same error, even though subfun is never even called. This could be a surprise: someone decides to add a nested function to working code, and it stops working because of an unrelated functionality using a dynamic regular expression within the same workspace. Ouch!

[Guillaume]

FWIW, I reported the issue to mathworks and it's been acknowledged. Whether or not it results in any change (at least in the documentation) is anybody guess. I wouldn't hold my breath...

[Walter Roberson]

I suspect you would see the same problem if you just had the "end" matching the "function" so you are using a static workspace,

   function test()
     regexp('123456','\d(?@$&)');
   end

Answer 2

[Stephen23]

The best and most general workaround I have found so far is to call regexp from a local function, with function handles to the nested function/s. Here is a complete working example.

function out = dynesty()
% Demonstrate calling a nested function from a dynamic regular expression.
str = '123456789';
out = {};
rgx = '\d(?@fDyn($&,$`,$''))'; % ($&=current,$`=precede,$'=follow)
localFun(@nestFun,str,rgx);
	function nestFun(varargin)
		out(end+1,1:nargin) = varargin;
	end
end
function varargout = localFun(fDyn,varargin) %#ok<INUSL>
[varargout{1:nargout}] = regexp(varargin{:});
end

Which returns exactly the expected output:

>> dynesty()
ans = 
    '1'    ''            '23456789'
    '2'    '1'           '3456789' 
    '3'    '12'          '456789'  
    '4'    '123'         '56789'   
    '5'    '1234'        '6789'    
    '6'    '12345'       '789'     
    '7'    '123456'      '89'      
    '8'    '1234567'     '9'       
    '9'    '12345678'    ''