Can RuntimeGeneratedFunctions.jl cause memory leak issue?

[vinhpb]

Hi,

I am writing a Genetic Programming-styled code to search for solutions to specific problems. The solutions are in the form of functions. Thus, I use RuntimeGeneratedFunctions.jl to generate functions in runtime in order to evaluate their fitnesses. As the code runs (on WSL) and time goes by, the amount of available RAM on my computer becomes less and less until the system forcibly closes the terminal. I suspect it is due to the generated functions. I wonder if it is a known problem and if there exists a solution. Thank you.

Here is the part of the code that involves RuntimeGeneratedFunctions.jl:

 function eval_solution(expr, data, eval_genfunc) 
      f = expr
      f1 = @RuntimeGeneratedFunction(f)
      fitness = evaluate_genfunc(f1, data)
      return fitness
 end

Here, expr is the Expr containing the content of the function to be generated, data is the data necessary to calculate the fitness of the generated function, eval_genfunc is a custom function to calculate the fitness of a generated function. eval_genfunc looks like this:

function eval_genfunc(f1, data) 
      parameters = f1(data)
      score = g(parameters)  % g performs a simulation with given parameters and extracts some information from there as the score
      return score
end

The function eval_solution( ) is used in multithreading mode in a main function:

function main(...)
 ...
 while iterate > 0
   Threads.@threads  for i in n_threads
      expr = ...  % Calling the function to generate an expr
      fitness = eval_solution(expr, data, eval_genfunc)
      ...
   end
   ...
   iterate -= 1
 end
 ...
end

[ChrisRackauckas]

The last word on this should be https://github.com/SciML/RuntimeGeneratedFunctions.jl/pull/63. @c42f is there still something that is retained, like the Julia compilation of the generated function?

[c42f]

IIUC, the answer is that yes, creating an unlimited number of functions at runtime will eventually cause your system to run out of memory.

Primarily I expect because all the old methods still exist in the Julia runtime: their internal CodeInfo objects, and the native code generated from that. I don't think there's any way to GC that stuff. Certainly it's not something we can control from RuntimeGeneratedFunctions - the best we can do here is to allow GC of the original Expr data structures (as pointed out - that's #63)