Re: atexit() stuff in thread-local reent?

[Sebastian Huber <sebastian dot huber at embedded-brains dot de>]

On 04/29/2013 11:26 PM, Jeff Johnston wrote:
> On 04/27/2013 07:52 AM, Sebastian Huber wrote:
> > Hello,
> >
> > currently the struct _reent contains some atexit() stuff:
> >
> > struct _reent {
> >    [...]
> >    /* atexit stuff */
> >    struct _atexit *_atexit;
> >    struct _atexit _atexit0;
> >    [...]
> > };
> >
> > on PowerPC for example this accounts for 400 of 1088 bytes of the
> > structure.  From my point of view the atexit() stuff belongs to the
> > global state and is not thread-specific.  I think all atexit() related
> > functions (in __call_atexit.c and  __atexit.c, are there more?) use
> > _GLOBAL_REENT.  This is related to the possibly unused _wrapup_reent()
> > function.  Maybe we should introduce two structures.  One for the global
> > state (contains thread-specific state + atexit() stuff) and one for the
> > thread-specific state.  This might save 400 bytes (37% of current
> > _reent) per thread.
>
> Instead, I might suggest that _atexit0 be removed from struct _reent and we
> simply declare a variable in reent/impure.c which is _global_atexit0 and add a
> define which is _GLOBAL_ATEXIT0.  The __atexit.c can be changed to use
> _GLOBAL_ATEXIT0 instead of _GLOBAL_REENT->_atexit0.

This is a good solution.  I will provide a patch for this in the next couple of 
days.

--
Sebastian Huber, embedded brains GmbH

Address : Dornierstr. 4, D-82178 Puchheim, Germany
Phone   : +49 89 189 47 41-16
Fax     : +49 89 189 47 41-09
E-Mail  : sebastian.huber@embedded-brains.de
PGP     : Public key available on request.

Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG.