Re: Help: correct way to handle struct return values.

[Kaz Kylheku <kaz at kylheku dot com>]

On 27.05.2017 20:14, Kaz Kylheku wrote:
> Hi all,
>
> In the current texinfo doc, the callback stub does this to
> return a value of type int:
>
>   /* Acts like puts with the file given at time of enclosure. */
>   void puts_binding(ffi_cif *cif, void *ret, void* args[],
>                   void *stream)
>   {
>     *(ffi_arg *)ret = fputs(*(char **)args[0], (FILE *)stream);
>   }
>
> Can someone show what the code would look like for a function
> that returns the following type, and work everywhere: all
> supported platforms, big or little endian:
>
>   struct little { char a, b };

This corresponds to libffi.call/struct5.c

Everything I'm seeing in the test suite seems like structs are handled
in the straighforward way in both ffi_call and closures. But
in this small struct test case there is this comment:

  /* This is a hack to get a properly aligned result buffer */
  test_structure_5 *ts5_result =
    (test_structure_5 *) malloc (sizeof(test_structure_5));


What are the alignment requirements here; must the small
structure have the same alignment as ffi_arg?

And why so, if its size is smaller? If this has to be,
say, 8 bytes aligned, doesn't it mean something will be writing
an 8 byte data unit into it? In that case, shouldn't
the malloc request be padded up to 8 so there is no overrun?

I am not concerned about the alignment because I'm using alloca
for the return value buffer given to ffi_call.

I am worried though about the size. alloca(2) could give me an
8 byte aligned pointer which is only two byte away from the next
object on the stack; if something writes 8 bytes there, that is
very bad.