Re: POINTER_ALIGNED_P (was: Re: CVS commit: src/sys)

[Valery Ushakov <uwe%stderr.spb.ru@localhost>]

On Tue, Feb 16, 2021 at 17:53:09 -0500, Christos Zoulas wrote:

> In this case "type" is a struct and we have __alignof() to handle
> it, but does this give the right answer?
> 
> Also ALIGNED_POINTER is not conditional to __NO_STRICT_ALIGNMENT and
> can be overriden (the opposite goes for POINTER_ALIGNED_P) I am all
> for having one macro, but how can we satisfy all the different
> semantics?

Well, it was you who did the definion of ALIGNED_POINTER centralized
and overridable :)

  revision 1.400
  date: 2012-01-25 00:03:36 +0400;  author: christos;  state: Exp;  lines: +26 -1;
  Use and define ALIGN() ALIGN_POINTER() and STACK_ALIGN() consistently,
  and avoid definining them in 10 different places if not needed.

ALIGNED_POINTER is overriden on x86 to be always true.  Surprisingly
it is not overriden for m68k and vax that are __NO_STRICT_ALIGNMENT.
That is most likely an oversight, but that will probably require some
cvs archaeology to confirm.  Some uses of ALIGNED_POINTER are inside
an __NO_STRICT_ALIGNMENT #ifdef.

We don't even seem to be sure about its semantics, as far as I can
tell (see bus space comments in my mail).

That's even more of a reason to stop doing aimless random changes
without getting some kind of understanding first.  The last thing we
need is ALIGNED_POINTER and POINTER_ALIGNED macros with slighly
different semantics both of which are counter-intuitive to begin with
(and riastradh@ even had to add a verbose comment for that).


-uwe