[RFC] Use %p conversion specifier to fix hppa compiler warning

[RFC] Use %p conversion specifier to fix hppa compiler warning

[Mark Kettenis]

Currently the compiler warns about hppa_pointer_to_address_hack().
Rather than adding ugly casts and making this an even worse hack, I
propose to remove it completely in favour of using the %p format
specifier.

In the past we didn't use the %p conversion specifier since it wasn't
portable; ancient UNIX didn't support it.  However, I'm confident that
we do not support those systems any more.  I've done some archeology,
and it seems %p is supported by Ultrix 4.0, HP-UX 10.01 and SunOS
4.1.3.

Unless people object, I'll commit the attached patch next weekend.
I'll also try to find all other %p workarounds in the code and get rid
of those.

Mark


Index: ChangeLog
from  Mark Kettenis  <kettenis@gnu.org>

	* hppa-tdep.c (hppa_pointer_to_address_hack): Remove function.
	(unwind_command): Use %p to print pointer.

Index: hppa-tdep.c
===================================================================
RCS file: /cvs/src/src/gdb/hppa-tdep.c,v
retrieving revision 1.207
diff -u -p -r1.207 hppa-tdep.c
--- hppa-tdep.c 12 Jun 2005 19:54:25 -0000 1.207
+++ hppa-tdep.c 12 Jun 2005 20:53:59 -0000
@@ -2402,16 +2402,6 @@ hppa_lookup_stub_minimal_symbol (const c
   return NULL;
 }
 
-/* Instead of this nasty cast, add a method pvoid() that prints out a
-   host VOID data type (remember %p isn't portable).  */
-
-static CORE_ADDR
-hppa_pointer_to_address_hack (void *ptr)
-{
-  gdb_assert (sizeof (ptr) == TYPE_LENGTH (builtin_type_void_data_ptr));
-  return POINTER_TO_ADDRESS (builtin_type_void_data_ptr, &ptr);
-}
-
 static void
 unwind_command (char *exp, int from_tty)
 {
@@ -2433,8 +2423,7 @@ unwind_command (char *exp, int from_tty)
       return;
     }
 
-  printf_unfiltered ("unwind_table_entry (0x%s):\n",
-		     paddr_nz (hppa_pointer_to_address_hack (u)));
+  printf_unfiltered ("unwind_table_entry (%p):\n", u);
 
   printf_unfiltered ("\tregion_start = ");
   print_address (u->region_start, gdb_stdout);

Re: [RFC] Use %p conversion specifier to fix hppa compiler warning

[Eli Zaretskii]

> Date: Sun, 12 Jun 2005 23:02:15 +0200 (CEST)
> From: Mark Kettenis <mark.kettenis@xs4all.nl>
> 
> Currently the compiler warns about hppa_pointer_to_address_hack().
> Rather than adding ugly casts and making this an even worse hack, I
> propose to remove it completely in favour of using the %p format
> specifier.
> 
> In the past we didn't use the %p conversion specifier since it wasn't
> portable; ancient UNIX didn't support it.  However, I'm confident that
> we do not support those systems any more.  I've done some archeology,
> and it seems %p is supported by Ultrix 4.0, HP-UX 10.01 and SunOS
> 4.1.3.

%p is ANSI/ISO C89, IIRC, so we can use it freely.  The one problem
with it is that its results are inconsistent: some libraries produce
"0x" before the address, others don't.  This is a minor inconsistency,
but with some numerical values, if you don't have the telltale "0x",
you may wonder whether the number is in hex or decimal.

Re: [RFC] Use %p conversion specifier to fix hppa compiler warning

[Mark Kettenis]

   Date: Mon, 13 Jun 2005 09:12:07 +0300
   From: Eli Zaretskii <eliz@gnu.org>

   > Date: Sun, 12 Jun 2005 23:02:15 +0200 (CEST)
   > From: Mark Kettenis <mark.kettenis@xs4all.nl>
   > 
   > Currently the compiler warns about hppa_pointer_to_address_hack().
   > Rather than adding ugly casts and making this an even worse hack, I
   > propose to remove it completely in favour of using the %p format
   > specifier.
   > 
   > In the past we didn't use the %p conversion specifier since it wasn't
   > portable; ancient UNIX didn't support it.  However, I'm confident that
   > we do not support those systems any more.  I've done some archeology,
   > and it seems %p is supported by Ultrix 4.0, HP-UX 10.01 and SunOS
   > 4.1.3.

   %p is ANSI/ISO C89, IIRC, so we can use it freely.  The one problem
   with it is that its results are inconsistent: some libraries produce
   "0x" before the address, others don't.  This is a minor inconsistency,
   but with some numerical values, if you don't have the telltale "0x",
   you may wonder whether the number is in hex or decimal.

Hmm, you're right.  Linux and *BSD do print the 0x, Solaris and HP-UX
don't.  Not sure if that's an issue, but I committed the attached
patch (which avoid %p) instead.  We use the %lx with cast to unsigned
long trick in several places.  Should be fine on ILP32 and LP64, and
the compiler will hopefully warn about it if we ever encounter a
systems that isn't one of those.

Mark

Index: ChangeLog
from  Mark Kettenis  <kettenis@gnu.org>

	* hppa-tdep.c (hppa_pointer_to_address_hack): Remove function.
	(unwind_command): Use %lx to print pointer.

Index: hppa-tdep.c
===================================================================
RCS file: /cvs/src/src/gdb/hppa-tdep.c,v
retrieving revision 1.209
diff -u -p -r1.209 hppa-tdep.c
--- hppa-tdep.c 18 Jun 2005 22:03:41 -0000 1.209
+++ hppa-tdep.c 19 Jun 2005 21:22:06 -0000
@@ -2395,16 +2395,6 @@ hppa_lookup_stub_minimal_symbol (const c
   return NULL;
 }
 
-/* Instead of this nasty cast, add a method pvoid() that prints out a
-   host VOID data type (remember %p isn't portable).  */
-
-static CORE_ADDR
-hppa_pointer_to_address_hack (void *ptr)
-{
-  gdb_assert (sizeof (ptr) == TYPE_LENGTH (builtin_type_void_data_ptr));
-  return POINTER_TO_ADDRESS (builtin_type_void_data_ptr, &ptr);
-}
-
 static void
 unwind_command (char *exp, int from_tty)
 {
@@ -2426,8 +2416,7 @@ unwind_command (char *exp, int from_tty)
       return;
     }
 
-  printf_unfiltered ("unwind_table_entry (0x%s):\n",
-		     paddr_nz (hppa_pointer_to_address_hack (u)));
+  printf_unfiltered ("unwind_table_entry (0x%lx):\n", (unsigned long)u);
 
   printf_unfiltered ("\tregion_start = ");
   print_address (u->region_start, gdb_stdout);