This is the mail archive of the newlib@sourceware.org mailing list for the newlib project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch] Move __libc_fini_array into a separate file.


I've made the register_fini function optional based on the _WANT_REGISTER_FINI flag. This flag can be set at configuration time via the newly added --enable-newlib-register-fini option. It can also be set manually if desired.

This should fix the problem and allow the new functionality where desired.

If there are any problems, please let me know.

2010-06-08 Jeff Johnston <jjohnstn@redhat.com>

        * configure.in: Add test for enable-newlib-register-fini
        to set the _WANT_REGISTER_FINI flag.
        * configure: Regenerated.
        * Makefile.in: Ditto.
        * newlib.hin: Add _WANT_REGISTER_FINI flag.
        * libc/stdlib/__call_atexit.c: Add test for
        _WANT_REGISTER_FINI to enable the register_fini function.

-- Jeff J.

On 06/08/2010 11:18 AM, Corinna Vinschen wrote:
On Jun 4 11:31, Jeff Johnston wrote:
On 06/03/2010 06:57 PM, Mark Mitchell wrote:
Jeff Johnston wrote:
What calls the new static routine register_fini??  I don't see anything
using it in this patch.

+static void
+register_fini(void) __attribute__((constructor (0)));

It's a constructor function.


Thanks,


Ah. Patch applied.

This patch breaks building Cygwin. When building from scratch, I'm getting this error message now:

   Creating library file: cygdll.a
   /ext/build/cygwin/src/i686-pc-cygwin/newlib/libc/libc.a(lib_a-__call_atexit.o):
   In function `register_fini':
   /home/corinna/src/cygwin/src/newlib/libc/stdlib/__call_atexit.c:49: undefined re
   ference to `__fini'
   collect2: ld returned 1 exit status

The problem is this.  In the former code, newlib only referenced _fini
if HAVE_INITFINI_ARRAY was defined.  If it wasn't defined, none of the
code in libc/misc/init.c was complied in.

Now, with the new code, _fini() is *always* referenced, even on systems
which don't define HAVE_INITFINI_ARRAY:

libc/stdlib/__call_atexit.c:

     static void
     register_fini(void)
     {
       if (&__libc_fini) {
     #ifdef HAVE_INITFINI_ARRAY
	extern void __libc_fini_array (void);
	atexit (__libc_fini_array);
     #else
	extern void _fini (void);
	atexit (_fini);
     #endif
       }
     }

AFAICS, the #else part of the conditional code just has to go away.

However, I'm not an expert in this stuff, so what do other's think
is the right solution?


Corinna


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]