Re: [PATCH] SPU Add support for nanosleep, sleep and usleep

[Jeff Johnston <jjohnstn at redhat dot com>]

Patrick Mansfield wrote:
> On Tue, Jun 12, 2007 at 03:14:20PM -0500, Joel Sherrill wrote:
> > Jeff Johnston wrote:
> > > Patrick Mansfield wrote:
>
> > > > Note the libc/include/sys/unistd.h: to change from:
> > > >    unsigned _EXFUN(usleep, (unsigned int __useconds));
> > > > To:
> > > >    int _EXFUN(usleep, (useconds_t __useconds));
> > > >
> > > > So we match both the existing usleep.c and POSIX.
> > > You can't just do that arbitrarily for Cygwin and RTEMS.  That 
> > > function prototype is protected by their OS flags being set on.  If 
> > > the prototype is not a typo error, then you must leave it alone.
> > The RTEMS implementation is
> >
> > int usleep(  useconds_t useconds );
> >
> > and the unistd.h we are using with the RPMs has it as:
> >
> > int _EXFUN(usleep, (useconds_t __useconds));
> >
> > Our patch to newlib has this fragment
> >
> > @@ -200,7 +200,7 @@
> > int    _EXFUN(getdtablesize, (void));
> > int    _EXFUN(setdtablesize, (int));
> > useconds_t _EXFUN(ualarm, (useconds_t __useconds, useconds_t __interval));
> > -unsigned _EXFUN(usleep, (unsigned int __useconds));
> > +int _EXFUN(usleep, (useconds_t __useconds));
> > #if !(defined  (_WINSOCK_H) || defined (__USE_W32_SOCKETS))
> > /* winsock[2].h defines as __stdcall, and with int as 2nd arg */
> >  int   _EXFUN(gethostname, (char *__name, size_t __len));
> >
> > So I guess we are using  the POSIX usleep profile and it just hasn't gotten
> > merged.
>
> > > You should put in a #else and then prototype usleep appropriately.  
>
> So OK to re-submit per Jeff's comment? Then Joel/cygwin will have to
> submit the change noted above.

Actually, I have confirmed with Chris Faylor that Cygwin also will use 
the correct prototytpe so I have moved the usleep prototype up to where 
the sleep prototype is and am ready to check in the modified patch.  I 
had to reorder the Makefile.am sources to make the patch apply.

> Jeff -
>
> Not for this patch, but should I just remove the #ifdef CYGWIN etc. if I
> am exposing an existing POSIX prototype?

Not sure what you mean by this.

> For new POSIX prototypes do I need an #ifdef? For example, I'm adding
> lockf() and defines for it, do you want an #ifdef __SPU__ around it?

For now, unless Cygwin and RTEMS want to use them too.

-- Jeff J.