Bug: Adding dummy loop breaks gradient in special circumstances

[guitargeek]

Simply adding a dummy loop with one iteration breaks certain gradients.

These minimal conditions are required to reproduce the problem:

1. A loop
2. Some "repacking" parameters from an array in a new array (commonly done in RooFit when forwarding subsets of parameters)
3. Doing some calculations with the repacked parameters in another function

My setup: Arch Linux with Clad master and clang 16.06.

I hope this problem can be fixed, as right now it is blocking me from implementing numeric integration routines in RooFit (similar to the reproducer, but the loop is not trivial).

Reproducer:

// Compile with clang++ -std=c++11 -fplugin=/usr/lib/clad.so cladIssue.cxx -o cladIssue
// Add the -DFIXED to see the fixed version

#include "clad/Differentiator/Differentiator.h"

#include <iostream>
#include <vector>

double func(double *params)
{
   return 1.0 / (params[0] * params[0]);
}

double wrapper(double *params)
{
   double out = 0.0;

#ifndef FIXED
   for (std::size_t i = 0; i < 1; ++i)
#else
#endif
   {
      double paramsPrime[] = {params[0]};
      out = out + func(paramsPrime);
   }

   return out;
}

int main()
{
   auto grad = clad::gradient(wrapper, "params");

   double x = 2.0;
   double result = 0.0;
   grad.execute(&x, &result);

   std::cout << result << std::endl;
}

[vaithak]

Hi @guitargeek,
There was an issue in the code generated for for-loops containing a function call with an array as a parameter.
I have verified that the above code works after the fix proposed in #640.