libiberty: Fix C89-isms in configure tests

Message ID 87h701jvk2.fsf@oldenburg.str.redhat.com
State Accepted
Headers
Series libiberty: Fix C89-isms in configure tests |

Checks

Context Check Description
snail/gcc-patch-check success Github commit url

Commit Message

Florian Weimer Oct. 18, 2022, 10:05 a.m. UTC
  libiberty/

	* acinclude.m4 (check for working strncmp): Add missing
	int return type and parameter list to the definition of main.
	Include <string.h> for string functions.  Avoid calling
	undeclared exit function.
        (stack direction for C alloca): Avoid calling undeclared exit
        function.
	* configure: Regenerate.

---
 libiberty/acinclude.m4 | 12 +++++++-----
 libiberty/configure    | 12 +++++++-----
 2 files changed, 14 insertions(+), 10 deletions(-)


base-commit: 54b316ff0d4f3bd823ad0b4d0011900948c5d40e
  

Comments

Jakub Jelinek Oct. 18, 2022, 12:57 p.m. UTC | #1
On Tue, Oct 18, 2022 at 12:05:49PM +0200, Florian Weimer via Gcc-patches wrote:
> libiberty/
> 
> 	* acinclude.m4 (check for working strncmp): Add missing
> 	int return type and parameter list to the definition of main.
> 	Include <string.h> for string functions.  Avoid calling
> 	undeclared exit function.
>         (stack direction for C alloca): Avoid calling undeclared exit
>         function.

Spaces instead of tabs.
I'd think we should #include <stdlib.h> for exit and keep exit, I vaguely
remember non-zero return from main doesn't always work reliably, which is
why e.g. in the testsuite we usually abort instead of return non-zero
from main.  Don't remember if it is just for some bare metal cases or
what, which on the either side probably don't have mmap.

	Jakub
  
Florian Weimer Oct. 18, 2022, 2:06 p.m. UTC | #2
* Jakub Jelinek:

> On Tue, Oct 18, 2022 at 12:05:49PM +0200, Florian Weimer via Gcc-patches wrote:
>> libiberty/
>> 
>> 	* acinclude.m4 (check for working strncmp): Add missing
>> 	int return type and parameter list to the definition of main.
>> 	Include <string.h> for string functions.  Avoid calling
>> 	undeclared exit function.
>>         (stack direction for C alloca): Avoid calling undeclared exit
>>         function.
>
> Spaces instead of tabs.

You mean I should use tabs throughout, right?

> I'd think we should #include <stdlib.h> for exit and keep exit, I vaguely
> remember non-zero return from main doesn't always work reliably, which is
> why e.g. in the testsuite we usually abort instead of return non-zero
> from main.  Don't remember if it is just for some bare metal cases or
> what, which on the either side probably don't have mmap.

Okay, will do that and send a v2.

By the way, the stack direction test currently gives incorrect results
on x86-64 due to -O2 and address comparison of unrelated objects.  I
assume this doesn't matter because we don't use it on compilers that
support alloca natively.

Thanks,
Florian
  
Jakub Jelinek Oct. 18, 2022, 2:08 p.m. UTC | #3
On Tue, Oct 18, 2022 at 04:06:17PM +0200, Florian Weimer wrote:
> By the way, the stack direction test currently gives incorrect results
> on x86-64 due to -O2 and address comparison of unrelated objects.  I
> assume this doesn't matter because we don't use it on compilers that
> support alloca natively.

Guess it would be better to cast the addresses to uintptr_t and
compare that.

	Jakub
  
Florian Weimer Oct. 18, 2022, 2:14 p.m. UTC | #4
* Jakub Jelinek:

> On Tue, Oct 18, 2022 at 04:06:17PM +0200, Florian Weimer wrote:
>> By the way, the stack direction test currently gives incorrect results
>> on x86-64 due to -O2 and address comparison of unrelated objects.  I
>> assume this doesn't matter because we don't use it on compilers that
>> support alloca natively.
>
> Guess it would be better to cast the addresses to uintptr_t and
> compare that.

But can we assume that uintptr_t is defined?  Or that <stdint.h> exists?

Thanks,
Florian
  
Jakub Jelinek Oct. 18, 2022, 2:19 p.m. UTC | #5
On Tue, Oct 18, 2022 at 04:14:53PM +0200, Florian Weimer wrote:
> * Jakub Jelinek:
> 
> > On Tue, Oct 18, 2022 at 04:06:17PM +0200, Florian Weimer wrote:
> >> By the way, the stack direction test currently gives incorrect results
> >> on x86-64 due to -O2 and address comparison of unrelated objects.  I
> >> assume this doesn't matter because we don't use it on compilers that
> >> support alloca natively.
> >
> > Guess it would be better to cast the addresses to uintptr_t and
> > compare that.
> 
> But can we assume that uintptr_t is defined?  Or that <stdint.h> exists?

In libiberty no, I'm afraid.

	Jakub
  

Patch

diff --git a/libiberty/acinclude.m4 b/libiberty/acinclude.m4
index 6bd127e9826..6e23ae906fc 100644
--- a/libiberty/acinclude.m4
+++ b/libiberty/acinclude.m4
@@ -24,6 +24,7 @@  AC_CACHE_CHECK([for working strncmp], ac_cv_func_strncmp_works,
 [AC_TRY_RUN([
 /* Test by Jim Wilson and Kaveh Ghazi.
    Check whether strncmp reads past the end of its string parameters. */
+#include <string.h>
 #include <sys/types.h>
 
 #ifdef HAVE_FCNTL_H
@@ -51,7 +52,8 @@  AC_CACHE_CHECK([for working strncmp], ac_cv_func_strncmp_works,
 
 #define MAP_LEN 0x10000
 
-main ()
+int
+main (void)
 {
 #if defined(HAVE_MMAP) || defined(HAVE_MMAP_ANYWHERE)
   char *p;
@@ -59,7 +61,7 @@  main ()
 
   dev_zero = open ("/dev/zero", O_RDONLY);
   if (dev_zero < 0)
-    exit (1);
+    return 1;
 
   p = (char *) mmap (0, MAP_LEN, PROT_READ|PROT_WRITE,
 		     MAP_ANON|MAP_PRIVATE, dev_zero, 0);
@@ -67,7 +69,7 @@  main ()
     p = (char *) mmap (0, MAP_LEN, PROT_READ|PROT_WRITE,
 		       MAP_ANON|MAP_PRIVATE, -1, 0);
   if (p == (char *)-1)
-    exit (2);
+    return 2;
   else
     {
       char *string = "__si_type_info";
@@ -79,7 +81,7 @@  main ()
       strncmp (r, q, 14);
     }
 #endif /* HAVE_MMAP || HAVE_MMAP_ANYWHERE */
-  exit (0);
+  return 0;
 }
 ], ac_cv_func_strncmp_works=yes, ac_cv_func_strncmp_works=no,
   ac_cv_func_strncmp_works=yes)
@@ -171,7 +173,7 @@  AC_CACHE_CHECK(stack direction for C alloca, ac_cv_c_stack_direction,
 }
 main ()
 {
-  exit (find_stack_direction() < 0);
+  return find_stack_direction() < 0;
 }],
   ac_cv_c_stack_direction=1,
   ac_cv_c_stack_direction=-1,
diff --git a/libiberty/configure b/libiberty/configure
index 65fc5002002..c871cc559ca 100755
--- a/libiberty/configure
+++ b/libiberty/configure
@@ -6798,7 +6798,7 @@  find_stack_direction ()
 }
 main ()
 {
-  exit (find_stack_direction() < 0);
+  return find_stack_direction() < 0;
 }
 _ACEOF
 if ac_fn_c_try_run "$LINENO"; then :
@@ -7621,6 +7621,7 @@  else
 
 /* Test by Jim Wilson and Kaveh Ghazi.
    Check whether strncmp reads past the end of its string parameters. */
+#include <string.h>
 #include <sys/types.h>
 
 #ifdef HAVE_FCNTL_H
@@ -7648,7 +7649,8 @@  else
 
 #define MAP_LEN 0x10000
 
-main ()
+int
+main (void)
 {
 #if defined(HAVE_MMAP) || defined(HAVE_MMAP_ANYWHERE)
   char *p;
@@ -7656,7 +7658,7 @@  main ()
 
   dev_zero = open ("/dev/zero", O_RDONLY);
   if (dev_zero < 0)
-    exit (1);
+    return 1;
 
   p = (char *) mmap (0, MAP_LEN, PROT_READ|PROT_WRITE,
 		     MAP_ANON|MAP_PRIVATE, dev_zero, 0);
@@ -7664,7 +7666,7 @@  main ()
     p = (char *) mmap (0, MAP_LEN, PROT_READ|PROT_WRITE,
 		       MAP_ANON|MAP_PRIVATE, -1, 0);
   if (p == (char *)-1)
-    exit (2);
+    return 2;
   else
     {
       char *string = "__si_type_info";
@@ -7676,7 +7678,7 @@  main ()
       strncmp (r, q, 14);
     }
 #endif /* HAVE_MMAP || HAVE_MMAP_ANYWHERE */
-  exit (0);
+  return 0;
 }
 
 _ACEOF