Don't add void prototypes on functions that would otherwise not need to be rewritten

[john-h-kastner]

This is a fix for an issue I observed while working on adding itypes to bodyless functions. A function that does not otherwise need to be rewritten will sometimes be rewritten in order to replace a missing prototype with (void). This was causing rewriting errors when an undefined function in an unwritable file was constrained to WILD in order to not add itypes on unwritable function, but it was still rewritten to use a void prototype. That particular issue doesn't exist on main prior to other change I made, but the test in this PR demonstrates another rewriting error that can be triggered.

The function is defined in a macro, so it cannot be rewritten without 3C raising an error. Constraints are generated so that the return type solve to WILD, but 3C still tries to rewrite it to insert a void prototype.

#define test_macro0 int *test_macro0()
test_macro0 {
  return 0;
}

This is fixed by only rewriting to insert the void prototype if the function already had to be rewritten for another reason.

[kyleheadley]

Checked C will not allow a function without a prototype within a checked scope. It was my understanding that this is why 3C adds the void to explicitly indicate the empty parameter list. Could you modify this to add it to all checked/safe functions?

[john-h-kastner]

Checked C will not allow a function without a prototype within a checked scope. It was my understanding that this is why 3C adds the void to explicitly indicate the empty parameter list. Could you modify this to add it to all checked/safe functions?

After these changes, 3C still adds a prototype for functions with checked or itype return types, so int *foo() { return 0; } will still convert to _Ptr<int> foo(void) { return 0; } as is required by Checked C.

CheckedC also requires that a functions has a prototype for it to be called from inside a checked scope, but this is not handled by 3C; see issue #382. The status of this issue has not changed.

[kyleheadley]

Thanks for the clarification about the other issue. I guess the tests added here will help that one to be more thorough once we work on it.

This change looks good.