Can the output of 'whos' operating on a workspace be duplicated for a structure array?

Hi guys -

I wrote a loop using 'whos' to filter out GUI figure elements when saving workspace variables so I'm not saving a copy of the GUI itself (the filter excludes all variables in which the 'class' is ui|graphics and/or gui handles).

What I'd like to do now is to replicate that filtering when operating on a structure array that contains different data types. So, is there a Matlab operation that works on a structure array the same way that 'whos' works on a workspace?

Specific technical answers welcome. Thanks!

[Jan]

What does this mean:

 zealots of "better coding practices" need not respond.

Are you aware that this kind of meta programming is a bad programming practice and you do not want to get a corresponding suggestion - e.g. simply to store all values to be exported in a struct, such that the export is trivial without any magic classification of variables? If you do not need answers like this, I'm leaving this thread. I will not help you to drill a hole in your knee.

[Jason]

All I mean by that is that I'm not interested in a philosophical discussion about preferred methodologies. I really do just want assistance on this specific technical question, knee-holes notwithstanding. What I'm interested in doing with my structure array is to filter out any data belonging to graphics or handles and to save that filtered structure array as a new array. I used 'whos' in the process of doing this before I generated this structure, but now instead of selectively condensing workspace variables into a structure, I need to weed out variables from a structure before I save it as another one.

[Jason]

Let me be more specific. In the course of saving workspace variables into an array, I filtered out GUI graphics info in the following way:

w = whos;
for a = 1:length(w) 
    Drive.(w(a).name) = eval(w(a).name); 
end
User = Drive; %Set of Drive variables to be edited by user through GUI
tosave1 = cellfun(@isempty, regexp({w.class}, '^matlab\.(ui|graphics)\.'));
tosave2 = cellfun(@isempty, regexp({w.name}, 'handles'));
tosave = logical(tosave1.*tosave2); 
save(matfilename, w(tosave).name);
MainGUI = getappdata(0, 'MainGUI');
setappdata(MainGUI,'User',User);

Later on I want to save a new version of User, but by this time there are once again GUI graphics elements in the array and I don't want them. I'm trying to remove them from User in a similar way as I removed them from the list of workspace variables when I first saved User.

[Walter Roberson]

What do you want the graphics objects and handles replaced with? The fields cannot just be deleted because you are using a structure array and any one field name might not have the same data types in all of the array elements.

[Jason]

I currently don't intend to replace the graphics data, just save the other structure variables in a new array. I'm having a recurring problem with GUI graphics data being saved in a structure such that when I load the structure in another section of my code that uses a different GUI, a saved version of the earlier GUI is opened too. That's what I'm trying to avoid.

[Walter Roberson]

Can you guarantee that in all elements of the structure array, that any given field name will always be graphic and/or handle over all of the array elements, with there never being a case where it is a graphic or handle in one array element but non-graphic or handle in some other array element? Including the possibility that it might be [] in some array elements but a graphic or handle in other elements of the array?

If you can make that guarantee then the list of fields to keep or drop can be determined by examining w(1) alone. If you cannot make that guarantee then all corresponding fields of w(:) have to be examined. Marking the field for keeping if even one non-graphic or handle is found? Marking the field for deletion if even one graphic or handle is found? Marking the field for deletion if all of the entries are either [] or graphic or handle, provided that at least one is not [] ?

[Stephen23]

See also Jason's earlier question:

https://www.mathworks.com/matlabcentral/answers/380011-how-to-pass-a-function-workspace-into-a-guidata-handles-structure-in-a-gui

[Jan]

@Jason: Your term "philosophical discussion" is polemical and not useful here. A better coding style would be the most efficient way to solve the problem.

Of course I know the problem of working in a brownfield project. You see already, that the current complicated code demands for even more complicated methods. This will increase the total complexity again and convert the brown field into a deep brown field.

[Jason]

@ Jan - Thank you again for your code sample - it pointed me in the direction I needed to go. Your "brown field" analogy is apt - I'm working with legacy code the best I can and working to make sure it doesn't become a deeper quagmire than it already is. Thank you for providing a technically helpful answer despite your misgivings.

[Jan]

@Jason: It seems like you are a experienced and responsible programmer, who can handle the dirty chain-saw tool I've provided. But newcomers in the forum might feel encouraged to use the evil eval methods and meta-programming also. Therefore Stephen's warnings and explanations are very important and useful.

@All readers: If you struggle with who, eval and other meta-programming methods, read Stephen's valuable Tutorial: Why and how to avoid eval.

Answer 1

[Jan]

UNTESTED CODE written in the forums interface:

function Sout = NoGUIFields(Sin)
% Input: Scalar struct
% Output: Input struct without fields, which contain '^matlab\.(ui|graphics)\.') classes
Field = fieldname(Sin);
Data  = struct2cell(Sin);
Sout  = struct();
for k = 1:numel(Field)
  if isa(Data{k}, 'struct')
    Sout.(Field{k}) = NoGUIField(Data{k});  % Recursive call
  elseif iscell(Data{k})
    Cin  = Data{k};
    Cout = cell(size(Cin));
    for iC = 1:numel(Cin)
      if isempty(regexp(class(Cin{k}), '^matlab\.(ui|graphics)\.')
        Cout{iC} = Cin{iC};
      end
    end
    Sout.(Field{k}) = Cout;
  elseif isempty(regexp(class(Data{k}, '^matlab\.(ui|graphics)\.')
    Sout.(Field{k}) = Sin.(Field{k});
  end
end
end

This removes all fields, which matches the '^matlab\.(ui|graphics)\.' filter. Substructs are scanned recursively. You can expand this for struct arrays easily on demand. I assume there are more cases to be considered, e.g. user defined handle classes. Therefore my estimation remains: This is a dirty programming style and it will cause more troubles than it solves.

By the way: cellfun('isempty', c) is much faster than cellfun(@isempty, c).

[Jason]

Thank you Jan, I'll take a look shortly and see what this can do. I appreciate your helpful and detailed response.