Defer fails to modify return values when a function has multiple return values about nelua-lang

[stefanos82]

I don't know whether you have seen it or not, but I have just found it and it's kind of baffling:

int nelua_main(int argc, char** argv) {
  {
    nelua_print_1(tmp_f(5));
    {
      nlmulret_nlint64_nlboolean _tmp1 = tmp_g(5);
      nelua_print_2(_tmp1.r1, _tmp1.r2);
    }
  }
  return 0;
}

Should not this code be

int nelua_main(int argc, char** argv) {
  {
    nelua_print_1(tmp_f(5));
  }
    {
      nlmulret_nlint64_nlboolean _tmp1 = tmp_g(5);
      nelua_print_2(_tmp1.r1, _tmp1.r2);
    }
  return 0;
}

because print(f(5)) and print(g(5)) are executed as two separate defers behind the scenes, correct?

[jrfondren]

The braces around nelua_print_2() are probably there to quickly free _tmp1 from occupying stack space - it's about multiple value returns from g() rather than anything getting deferred.

[stefanos82]

OK, I have found the bug; the generated C code for defer takes place in wrong line(s):

/* Buggy version */
nlmulret_nlint64_nlboolean tmp_g(int64_t n) {
  nlmulret_nlint64_nlboolean _mulret_1;
  _mulret_1.r1 = n;
  _mulret_1.r2 = true;
  { /* defer */
    n = (n + 1);
  }
  return _mulret_1;
}

It should be

/* Fixed version*/
nlmulret_nlint64_nlboolean tmp_g(int64_t n) {
  nlmulret_nlint64_nlboolean _mulret_1;
  { /* defer */
    n = (n + 1);
  }
  _mulret_1.r1 = n;
  _mulret_1.r2 = true;
  return _mulret_1;
}

Update: After thinking about it, my fix violates the logic of defer mechanism that is supposed to take place as last. I wonder how @edubart could solve this peculiar case...maybe with the use of GOTO labels? 🤔

Update 2: On second thought, the code below should work too:

nlmulret_nlint64_nlboolean tmp_g(int64_t n) {
  nlmulret_nlint64_nlboolean _mulret_1;
  { /* defer */
    n = (n + 1);
  }
  return _mulret_1  = (nlmulret_nlint64_nlboolean){.r1=n, .r2=true};
}

Since we are deferring n only and then return n, true, a compound literal should work as shown above.

[edubart]

This is a serious bug, I try to fix it next weekend. Thanks for reporting!

[stefanos82]

@edubart keep up the good work brother 👍

@jrfondren Good job my friend with your thorough investigation; that makes us two now! 😃

[edubart]

I fixed the misbehavior in dc479bf, by definition defer should only execute after evaluating all function statements, including return expressions.