How can a breakpoint affect the result of a computation?

I am debugging and I have a block of code that gives a different answer depending on the location of a breakpoint. The code is included here:

%---compute c1hat---%
c1hat = zeros(size(c));
for i = ind_compute.'
    %---make function file---%
    fid = fopen(['intgrnd_',funcname,'.m'],'w');
    fprintf(fid, '%s \n\n', ['function f = intgrnd_',funcname,'(w,t,p,c)']);
    fprintf(fid, '%s \n', ['f = (',funcname,'_hat(w,t,p,c)).*conj(',Zstr{i},').*(0.5).*(sin(w/2).^2).*sin(t);']);
    fclose(fid);

      %----make function handle---%
      integrand = eval(['@(w,t,p) intgrnd_',funcname,'(w,t,p,c);']);

      %---perform integration---%
      c1hat(i) = simp3D(integrand,0,2*pi,0,pi,0,2*pi,32,32,32);
  end

essentially the code just writes a .m function file; makes a function handle to call it with a particular argument, c; and then uses the function handle to perform numerical integration.

If I put a breakpoint on the line that performs the integration and then manually step through each iteration of the for loop by pressing F5 over and over again it gives me the right answer (stored in the variable chat). If instead I put a breakpoint after the for loop to check what is stored in chat I find the wrong answer: specifically, all the elements of chat are the same and equal to the value from the first iteration.

How is it possible that the location of a breakpoint while debugging could alter the results of a computation? And how do I fix this.

Thank you very much for your time!

Answer 1

[Jan]

This is the expected behaviour. If you create an M-file dynamically, it is not reloaded during normal program execution. Therefore this technique is not recommended.

You can force the reloading by

clear(['intgrnd_', funcname]);

Answer 2

[Walter Roberson]

Bad behavior such as this is common when you use eval()

I am not certain that I followed everything, but it appears likely to me that you can avoid these troubles by using anonymous functions, perhaps in conjunction with str2func() . At the moment, I do not see any reason why intgrnd_* needs to be written to a file.

Note: in newer versions, str2func knows anonymous functions syntax.

[Oliver]

The reason for writing the .m file is for speed. The function is an expansion of harmonic functions so it has many terms. I tried just combining anonymous functions (for example: f = @(x) a*g(x)+b*h(x)), but it is very slow compared to an explicitly written .m file. The integration is finding coefficients for the expansion.

[Oliver]

Any other ideas for creating a series expansion that can be called as a function, and whose coefficients change dynamically?