Re: gcc-2.95 OK, gcc-{3,4}.X not OK

[Cedric Roux <cedric dot roux at acri-st dot fr>]

Andris Kalnozols wrote:
> There are no coding nor compilation errors with either
>
>   pcptr = pcptr->code = nop;
>     or
>   pcptr->code = pcptr = nop;
>
> As long as the pointers are actually pointing to something,
> this is typical linked list processing.

Well, this is debatable.
If by:
  pcptr->code = pcptr = nop;
you mean:
  pcptr = nop;
  pcptr->code = nop;
(assuming nop == NULL)
then in the second assignment pcptr is NULL,
you dereference a NULL pointer. (And you're
lucky you get a segmentation fault, on some
architectures you have no crash at all and
left with a funky bug.)

>         pcptr->code = fnc_A();

What does that mean in your head?
fnc_A modifies pcptr, so you expect
pcptr in pcptr->code to be the value
before the call or after?

In a hypothetical assembly language, your statement could be:

  load  (pcptr), reg1
  add   #offset(code in PC), reg1
  call  fnc_A
  store regret, (reg1)

Or do you expect:

  call  fnc_A
  load  (pcptr), reg1
  add   #offset(code in PC), reg1
  store regret, (reg1)

This is quite different. I don't know what gcc does, but
I know your pattern is weird and I guess you
should avoid it in production code. It will
give some headaches to those who will need
to maintain the code.

Cédric.