Difference between work vector size in generated header and c files

[andreasbell]

For the generated code there is a discrepancy between the work vector size (sz_w) between the .h file and .c file.

Example:

a = ca.SX.sym("a", 1)
fun = ca.Function("fun", [a], [a*a])
fun.generate("fun.c", {"with_header":True})

For the generated .c file we get the correct work vector size

CASADI_SYMBOL_EXPORT int fun_work(casadi_int *sz_arg, casadi_int* sz_res, casadi_int *sz_iw, casadi_int *sz_w) {
  if (sz_arg) *sz_arg = 1;
  if (sz_res) *sz_res = 1;
  if (sz_iw) *sz_iw = 0;
  if (sz_w) *sz_w = 0;
  return 0;
}

However for the .h file we get

#define fun_SZ_W 1

NOTE: When using the code generation option {"avoid_stack": True} we get the same work vector size for both the .c and .h file.

[jaeandersson]

Too be clear, the risk here is that you allocate slightly too much memory (work vector size 1 instead of 0 which is what is needed).