Re: [patch]: new returns misaligned pointers

[Richard Henderson <rth at cygnus dot com>]

On Fri, Nov 12, 1999 at 03:52:37PM -0800, Mark Mitchell wrote:
> Makes sense.  Could you mention for which ports this makes a
> difference?  If it does make a difference with an existing port...

Of the existing live ports, 

  arm/arm.h:#define BIGGEST_ALIGNMENT  32
  arm/thumb.h:#define BIGGEST_ALIGNMENT 32
  c4x/c4x.h:#define BIGGEST_ALIGNMENT     32
  fr30/fr30.h:#define BIGGEST_ALIGNMENT 32
  i386/cygwin.h:#define BIGGEST_ALIGNMENT 128
  i960/i960.h:#define BIGGEST_ALIGNMENT 128
  m32r/m32r.h:#define BIGGEST_ALIGNMENT 32
  m68k/news.h:#define BIGGEST_ALIGNMENT 32
  mn10200/mn10200.h:#define BIGGEST_ALIGNMENT     16
  mn10300/mn10300.h:#define BIGGEST_ALIGNMENT     32
  ns32k/ns32k.h:#define BIGGEST_ALIGNMENT 32
  rs6000/sysv4.h:#define BIGGEST_ALIGNMENT ((TARGET_EABI) ? 64 : 128)
  sh/sh.h:#define BIGGEST_ALIGNMENT  (TARGET_ALIGN_DOUBLE ? 64 : 32)
  sparc/sparc.h:#define BIGGEST_ALIGNMENT (TARGET_ARCH64 ? 128 : 64)
  v850/v850.h:#define BIGGEST_ALIGNMENT   32
  vax/vax.h:#define BIGGEST_ALIGNMENT 32

Of the 16, 11 have BIGGEST_ALIGNMENT smaller than `double', and
4 have BIGGEST_ALIGNMENT larger, for a 128-bit `long double'.
SH has special problems due to -malign-double affecting things.

I hadn't realized there were quite so many victims of change.

It's up to someone C++-ish to decide if breaking the ABI for
those defining BIGGEST_ALIGNMENT smaller than `double' is a
reasonable thing to do.  For the four ports that have a large
`long double' things are clearly already broken.

Might I suggest, if we're going to decide to break things, 
that instead of changing to always using BIGGEST_ALIGNMENT,
that we use the alignment of the type being allocated?  Seems
like that way we'd avoid wasting little bits of memory
unnecessarily.


r~