Re: [PATCH 6/7] hw/isa: Assert isa_register_portio_list() gets non-NULL ISA device

[Philippe Mathieu-Daudé]

On 8/2/23 20:47, Richard Henderson wrote:
> On 2/7/23 14:07, Philippe Mathieu-Daudé wrote:
> > The previous commit removed the single call to
> > isa_register_portio_list() with dev=NULL. To be
> > sure we won't reintroduce such weird (ab)use,
> > add an assertion.
> >
> > Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
>
> Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
>
> I wonder how much use of __attribute__((nonnull)) we should be making.

__attribute__((nonnull)) is compile-time, but seems weaker than the
good old runtime assert():

 void a0(void *ptr)
 {
   assert(ptr);
 }

 __attribute__((nonnull)) void a1(void *ptr)
 {
   // can no use assert(ptr) because compiler warning
 }

 void b0(void *x)
 {
   a(NULL); // runtime assertion
 }

 void b(void *x)
 {
   a1(NULL); // compile error
 }

 void c0(void *x)
 {
   a0(x);
 }

 void c1(void *x)
 {
   a1(x);
 }

 void d0(void *x)
 {
   c0(NULL); // runtime assertion
 }

 void d1(void *x)
 {
   c1(NULL); // no compile error, no assertion!
 }

> I realize we'd probably want to add -fno-delete-null-pointer-checks if 
> we make too much use of that.