Re: Is there a HOST_BITS_PER_POINTER macro?

Re: Is there a HOST_BITS_PER_POINTER macro?

[Kaveh R. Ghazi]

 > From: Richard Henderson <rth@dot.cygnus.com>
 > 
 > On Thu, Feb 05, 1998 at 07:45:09PM -0500, Kaveh R. Ghazi wrote:
 > > I get a warning from "-W -Wall" which complains about int format vs
 > > pointer argument.  Is there a way to define a macro in machmode.h
 > > which can cast the pointer to the appropriately sized integer type?
 >  
 > Ideally you would use "%p" everywhere, but to support bootstrapping,
 > we should probably autoconf a test for that.
 > r~

	Good point.  I guess it would have to be a runtime check of the
*printf functions in libc.  How about using code like this?


 > #include <stdio.h>
 > main()
 > {
 >   char foo[30];
 >   char * ptr = (char *) 0;
 > 
 >   sprintf(foo, "%p", ptr)
 >   exit (strcmp(foo, "0"));
 > }


and then doing AC_DEFINE(HOST_PTR_PRINTF, "%p") if it passes?
(machmode.h won't define it if its already defined.)

	If you think the test is sufficient, I'll whip up a patch.

		--Kaveh
--
Kaveh R. Ghazi			Project Manager / Custom Development
ghazi@caip.rutgers.edu		ICon CMT Corp.

Re: Is there a HOST_BITS_PER_POINTER macro?

[Richard Henderson]

On Fri, Feb 06, 1998 at 11:48:08AM -0500, Kaveh R. Ghazi wrote:
>  >   sprintf(foo, "%p", ptr)
>  >   exit (strcmp(foo, "0"));

The problem is that it is implementation defined what
format pointers are represented in.  So you might easily
get any of

	0
	00000000
	0x0
	0x00000000
	0000:0000

I can think of two possible solutions:

	char buf[64];
	sprintf (buf, "%p", buf);
#ifdef HAVE_INDEX
	exit (index (buf, 'p') == NULL);
#else
	exit (strchr (buf, 'p') == NULL);
#endif

since it seems unlikely that any representation will result in 'p',
though of course that's not the only failure mode, or
	
	char buf[64];
	char *p = buf, *q = NULL;
	sprintf(buf, "%p", p);
	sscanf(buf, "%p", &q);
	exit (p != q);

if sscanf can be trusted to read them back in.  I would think that's
required, but...


r~

Re: Is there a HOST_BITS_PER_POINTER macro?

[Richard Henderson]

On Sat, Feb 07, 1998 at 04:09:17PM -0500, Kaveh R. Ghazi wrote:
> 	I like your second solution better.  If sscanf doesn't work,
> gcc will fall back on the definitions in machmode.h.  Here's a patch
> to implement this.  If you think its okay, would you please install it?

Done.


r~

Re: Is there a HOST_BITS_PER_POINTER macro?

[Kaveh R. Ghazi]

 > From: Richard Henderson <rth@dot.cygnus.com>
 > 
 > The problem is that it is implementation defined what
 > format pointers are represented in.  So you might easily
 > get any of
 > 
 > 	0
 > 	00000000
 > 	0x0
 > 	0x00000000
 > 	0000:0000
 > 
 > I can think of two possible solutions:
 > 
 > [...]
 > 
 > or 
 > 	
 > 	char buf[64];
 > 	char *p = buf, *q = NULL;
 > 	sprintf(buf, "%p", p);
 > 	sscanf(buf, "%p", &q);
 > 	exit (p != q);
 > 
 > if sscanf can be trusted to read them back in.  I would think that's
 > required, but...
 > r~

	I like your second solution better.  If sscanf doesn't work,
gcc will fall back on the definitions in machmode.h.  Here's a patch
to implement this.  If you think its okay, would you please install it?

		Thanks,
		--Kaveh


Sat Feb  7 15:11:28 1998  Kaveh R. Ghazi  <ghazi@caip.rutgers.edu>

        * aclocal.m4 (GCC_FUNC_PRINTF_PTR): New macro to test the printf
        functions for whether they support the %p format specifier.

        * acconfig.h (HOST_PTR_PRINTF): Insert stub for autoconf.

        * configure.in (GCC_FUNC_PRINTF_PTR): Use it.


diff -r -u orig/egcs-980205/gcc/acconfig.h egcs-980205/gcc/acconfig.h
--- orig/egcs-980205/gcc/acconfig.h	Sat Jan 31 19:48:00 1998
+++ egcs-980205/gcc/acconfig.h	Sat Feb  7 14:27:17 1998
@@ -1,3 +1,6 @@
+/* Define to "%p" if printf supports it, else machmode.h will define it.  */
+#undef HOST_PTR_PRINTF
+
 /* Define if your cpp understands the stringify operator.  */
 #undef HAVE_CPP_STRINGIFY
 
diff -r -u orig/egcs-980205/gcc/aclocal.m4 egcs-980205/gcc/aclocal.m4
--- orig/egcs-980205/gcc/aclocal.m4	Sat Jan 31 19:48:01 1998
+++ egcs-980205/gcc/aclocal.m4	Sat Feb  7 14:33:34 1998
@@ -42,6 +42,27 @@
 done
 ])
 
+dnl See if the printf functions in libc support %p in format strings.
+AC_DEFUN(GCC_FUNC_PRINTF_PTR,
+[AC_CACHE_CHECK(whether the printf functions support %p,
+  gcc_cv_func_printf_ptr,
+[AC_TRY_RUN([#include <stdio.h>
+
+main()
+{
+  char buf[64];
+  char *p = buf, *q = NULL;
+  sprintf(buf, "%p", p);
+  sscanf(buf, "%p", &q);
+  exit (p != q);
+}], gcc_cv_func_printf_ptr=yes, gcc_cv_func_printf_ptr=no,
+	gcc_cv_func_printf_ptr=no)
+rm -f core core.* *.core])
+if test $gcc_cv_func_printf_ptr = yes ; then
+  AC_DEFINE(HOST_PTR_PRINTF, "%p")
+fi
+])
+
 dnl See if symbolic links work and if not, try to substitute either hard links or simple copy.
 AC_DEFUN(GCC_PROG_LN_S,
 [AC_MSG_CHECKING(whether ln -s works)
diff -r -u orig/egcs-980205/gcc/configure.in egcs-980205/gcc/configure.in
--- orig/egcs-980205/gcc/configure.in	Sat Jan 31 19:48:05 1998
+++ egcs-980205/gcc/configure.in	Sat Feb  7 14:25:13 1998
@@ -180,6 +180,8 @@
 AC_CHECK_FUNCS(strtoul bsearch strerror putenv popen vprintf bcopy bzero bcmp \
 	index rindex kill getrlimit setrlimit atoll atoq)
 
+GCC_FUNC_PRINTF_PTR
+
 GCC_NEED_DECLARATIONS(malloc realloc calloc free bcopy bzero bcmp \
 	index rindex getenv atol sbrk abort)
 
--
Kaveh R. Ghazi			Project Manager / Custom Development
ghazi@caip.rutgers.edu		ICon CMT Corp.

Re: Is there a HOST_BITS_PER_POINTER macro?

[Jeffrey A Law]

  In message < 19980205221028.48772@dot.cygnus.com >you write:
  > On Thu, Feb 05, 1998 at 07:45:09PM -0500, Kaveh R. Ghazi wrote:
  > > I get a warning from "-W -Wall" which complains about int format vs
  > > pointer argument.  Is there a way to define a macro in machmode.h
  > > which can cast the pointer to the appropriately sized integer type?
  > 
  > Ideally you would use "%p" everywhere, but to support bootstrapping,
  > we should probably autoconf a test for that.
Agreed.

We autoconf test for it, then define/use HOST_PTR_PRINTF appropriately.

jeff

Re: Is there a HOST_BITS_PER_POINTER macro?

[Joern Rennecke]

> If I have the following in gcc's src code:
> 
>   char * foo;
>   [...]
>   printf(HOST_PTR_PRINTF, foo);
> 
> I get a warning from "-W -Wall" which complains about int format vs
> pointer argument.  Is there a way to define a macro in machmode.h
> which can cast the pointer to the appropriately sized integer type?
> I'd like to do the following 

Even if you had something like that (e.g. ptrdiff_t on systems that
have it), that would be little help: You'd have to make sure that the
integer type matches the format specifier.

I think the most portable you can get without autoconf help is to
use long consistently, i.e. cast to long and use a long format specifier.

Of course that will still fail when pointers are wider than long...

Re: Is there a HOST_BITS_PER_POINTER macro?

[Richard Henderson]

On Thu, Feb 05, 1998 at 07:45:09PM -0500, Kaveh R. Ghazi wrote:
> I get a warning from "-W -Wall" which complains about int format vs
> pointer argument.  Is there a way to define a macro in machmode.h
> which can cast the pointer to the appropriately sized integer type?

Ideally you would use "%p" everywhere, but to support bootstrapping,
we should probably autoconf a test for that.



r~

Is there a HOST_BITS_PER_POINTER macro?

[Kaveh R. Ghazi]

If I have the following in gcc's src code:

  char * foo;
  [...]
  printf(HOST_PTR_PRINTF, foo);

I get a warning from "-W -Wall" which complains about int format vs
pointer argument.  Is there a way to define a macro in machmode.h
which can cast the pointer to the appropriately sized integer type?
I'd like to do the following 

  printf(HOST_PTR_PRINTF, (HOST_PTR_PRINTF_CAST) foo);

I think I need something that can be evaluated by cpp, not cc1.  So that
means I can't use sizeof.  If I could find a HOST_BITS_PER_PTR macro, I
could handle this in machmode.h.  But I can't find something like this
in the config files. 

		Thanks,
		--Kaveh
--
Kaveh R. Ghazi			Project Manager / Custom Development
ghazi@caip.rutgers.edu		ICon CMT Corp.