Couple of bugs

[Christian Ludlam]

On 21 Sep Tom Hughes wrote:

> In message <e5d2bd254f.Jo at hobbes.bass-software.com>
>         John Tytgat <John.Tytgat at aaug.net> wrote:
> 
> > It does.  The "function" type is also used in the PCI swi interface file
> > but it is not clear to me if this is the same APCS interface as for
> > DragAnAobject swi.
> 
> The problem is that DefMod doesn't have a syntax to fully describe
> functions and their arguments so it can't spit out a proper prototype
> and instead the function type in DefMod is just mapped to void * in
> the C output I think.
> 
> A lot of the function points taken by SWIs are not very amenable to
> being represented as C prototypes anyway I suspect...

I don't think it would need to describe the prototype fully - currently you
need to cast your function, but even with an EXPLICIT cast Norcroft gives you
a "cast between function and non-function types" warning.

If it was prototyped as "void (*)(void)" then you'd still need the cast, but
you'd get rid of the warning.

I did this when I modified DefMod to generate veneers to let me build Zap
modes in C. The code changed (in cheader.c)

      case def_TYPE_ASM:
         if ((rc = fprintf (file, "void %s", v)) < 0)
            goto finish; /*v != NULL*/
      break;

into this:

      case def_TYPE_ASM:
         if ((rc = fprintf (file, "void (*%s)(void)", v)) < 0)
            goto finish; /*v != NULL*/
      break;

We'd probably need to go through all the swi files that use .Asm and check
that they really do mean function pointers, and not just (void *).

-- 
Christian Ludlam
christian at recoil.org