Prevent figure call from .p file

I need to run a function from a software package that is written in a .p file. To run it on a cluster, I need to start matlab with the following options:

matlab -nojvm -nodesktop -nosplash -nodisplay -singleCompThread

The .p file attempts to make figures which cause matlab to crash due to nojvm. Is there anyway to prevent the figure creation without being able to edit the source code?

Thanks in advance!

[Rik]

It should not work, but you can try something like this:

function x=figure(varargin), x = 1+rand; end

[Fangjun Jiang]

I doubt this would work. What about the usual plot() commands after the figure()? But you could try and report what happened.

[Rik]

It depends on how and when the p file determines which function to run. I personally think it should give preference to built-in functions, but I never tested it. This seems the easiest attack vector other than modifying the actual internal function (which is probably very much non-trivial in the case of figure).

[Oliver Warrington]

Thanks for the suggestion @Rik. No luck though I'm afraid. Seems there is preference for built-in functions. Unfortunate for this case but I'm sure reasonable overall.

[Walter Roberson]

When I test with my own .p and my own replacement figure(), the replacement does get called.

S = "function test_pcode; fprintf('function called!\n'); fig = figure(); dfig = double(fig); fprintf('figure number was %d\n', dfig); end"

fid = fopen('test_pcode.m', 'w');
fwrite(fid, S);
fclose(fid)

which test_pcode

pcode test_pcode

which test_pcode
clear figure
which figure

test_pcode

function fignum = figure(varargin)
  fprintf('replacement figure called!\n');
  fignum = randi(20);
end

When I test this code here in Answers, the replacement figure message does not get displayed -- it is calling the built-in figure() it appears. But when I test the code in R2022b rerelease on my desktop, the message does appear indicating that my replacement function was called.

[Rik]

The whole point of p code is to prevent the end user digging around, so perhaps the writer of the code used the builtin() function to ensure built-in functions are called.

The inconsistent behavior is odd. I wonder whether it has something to do with the OS.

[Walter Roberson]

My guess at the moment would be that it has to do with Answers running in a LiveScript environment. LiveScript must be implicitly intercepting figure() so that it can inline figures.

Answer 1

[Oliver Warrington]

I did get this to work by using a brute force version of replacing the built in functions, as suggested in the comments.

I created dummy versions of the graphics, graph2d, graph3d and specgraph folders. Replaced all the functions, so no subsequent plot call of any kind would throw an error. Then removed the built in versions from the path and added my dummy versions.

    rmpath(genpath('/usr/local/MATLAB/R2021b/toolbox/matlab/graph2d'))
    rmpath(genpath('/usr/local/MATLAB/R2021b/toolbox/matlab/graphics'))
    rmpath(genpath('/usr/local/MATLAB/R2021b/toolbox/matlab/specgraph'))
    rmpath(genpath('/usr/local/MATLAB/R2021b/toolbox/matlab/graph3d'))
    addpath(genpath('./graphics'));
    addpath(genpath('./graph2d'));
    addpath(genpath('./specgraph'));
    addpath(genpath('./graph3d'));

Not sure this would be recommended in general, but my results are as expected and I can now run it on a cluster!

[Rik]

I'm sure I would not recommend this in general, but in your case it seems the only feasible solution.