Anonymous function context being destroyed during save/load cycle into a mat file.

I'm creating a function handle in the code shown here, beneath the red lines:

- the anonymous function in the cell array takes 3 args, and passes them through to online_rpca_model_script2 with the argument parameter_instance already closed/closured by the outer anonymous function ( exp_f_handle ).

This at some point is saved into a .mat file. Before it is saved, I did a little jump into debug mode to poke around:

So the variable worker_args is the cell array that was created underneath the red lines in the first image. The worker_args{2} is the function handle in question, and I get the metadata for that function handle by calling "functions()" on it. I then look at all the element of the workspace field of the returned structure. We can see that everything that was in scope during the anonymous function's creation is there. Particularly, parameter_instance is in the 2nd element of the workspace field. I then save worker_args to a .mat file.

I then load this same .mat file in a second Matlab instance, which has exactly the same search path as the first Matlab instance. They are literally launched from the same directory, and I've run the same addpath script in both.

Then I call functions() on worker_args{2}, which is our function handle in question. We can see from the returned struct that the function strings match, the files match (you can't see that from the images, but they do), however: the workspace field has changed.

In particular, workspace{2} has lost the parameter_instance variable, workspace{3} has lost condor_obj and experiment_opts, and workspace{4} has lost a whole load of stuff. So now, when I call the function handle in worker_args{2}, it complains that the variable parameter_instance is undefined, despite that fact that it definitely was defined and closured according to the second image!

So something in the save-load process is screwing up my function handle, and I would like to know what I'm doing wrong, what I'm wrongly assuming, or what the bug is.

[Matt J]

We cannot see from what you posted where you introduced these different variables into the scope of the anonymous function. Any external variables that the function will make use of must be assigned their relevant values before the anonymous function is defined. These will be the values the function handle snapshots and will normally be held in workspace{1}. Therefore, it is suspicious that your original workspace{1} is empty.

Also, we need to know whether the 2 MATLAB instances are the same version. The behavior of the FUNCTIONS command is not guaranteed consistent across MATLAB versions.

[Sanj B]

The two MATLAB instances are both 2015a.

So, I'm not sure how anon-functions work in MATLAB, but circumstantially, I would think workspace{1} would only get filled if tmpY1, tmpY2, and tmpopts, were closured.

I say this because parameter_instance is closed in the next outer scope and appears in workspace{2}. The vars in workspace{3} are introduced in the next outer scope after that. Hence because tmpY1, tmpY2, and tmpopts are arguments to the function handle and are not closed, there is nothing to put in workspace{1}.

Even if that is suspicious, it doesn't explain why stuff gets wiped from the anon-functions context after a save-load.

[Matt J]

I don't quite know what you mean by "next outer scope". You mean the anonymous function is enclosed in a series of nested functions?

it doesn't explain why stuff gets wiped from the anon-functions context after a save-load.

Well, it might. There is sometimes a lot of useless data in the workspace of an anonymous function, stuff that never gets used and just serves to consume memory when you save the anonymous function to disk or export it elsewhere. This FEX file was my effort to try and clean this junk away,

http://www.mathworks.com/matlabcentral/fileexchange/45749-clean-anonymous-functions-of-excess-memory

It may be that R2015 introduced a similar attempt to do this, maybe not so successfully.