[Bug c/46859] New: Attribute depends on location

[Bug c/46859] New: Attribute depends on location

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46859

           Summary: Attribute depends on location
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: hjl.tools@gmail.com


foo2 in this test isn't aligned:

[hjl@gnu-6 pages-1]$ cat a.c
 __attribute__((aligned(4096))) 
int *
foo1 ()
{
  return 0;
}

int *
 __attribute__((aligned(4096))) 
foo2 ()
{
  return 0;
}

 __attribute__((aligned(4096))) 
int
foo3 ()
{
  return 0;
}

int
 __attribute__((aligned(4096))) 
foo4 ()
{
  return 0;
}
[hjl@gnu-6 pages-1]$ gcc -S -O2 a.c
[hjl@gnu-6 pages-1]$ cat a.s
    .file    "a.c"
    .text
    .align 4096
.globl foo1
    .type    foo1, @function
foo1:
.LFB0:
    .cfi_startproc
    xorl    %eax, %eax
    ret
    .cfi_endproc
.LFE0:
    .size    foo1, .-foo1
    .p2align 4,,15
.globl foo2
    .type    foo2, @function
foo2:
.LFB1:
    .cfi_startproc
    xorl    %eax, %eax
    ret
    .cfi_endproc
.LFE1:
    .size    foo2, .-foo2
    .align 4096
.globl foo3
    .type    foo3, @function
foo3:
.LFB2:
    .cfi_startproc
    xorl    %eax, %eax
    ret
    .cfi_endproc
.LFE2:
    .size    foo3, .-foo3
    .align 4096
.globl foo4
    .type    foo4, @function
foo4:
.LFB3:
    .cfi_startproc
    xorl    %eax, %eax
    ret
    .cfi_endproc
.LFE3:
    .size    foo4, .-foo4

[Bug c/46859] Attribute depends on location

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46859

--- Comment #1 from H.J. Lu <hjl.tools at gmail dot com> 2010-12-09 15:44:49 UTC ---
Shouldn't foo1/foo2 have the same alignment?

[hjl@gnu-6 pages-1]$ cat x.c
typedef int * ptr_t;

ptr_t
 __attribute__((aligned(4096))) 
foo1 ()
{
  return 0;
}

int *
 __attribute__((aligned(4096))) 
foo2 ()
{
  return 0;
}
[hjl@gnu-6 pages-1]$ gcc -S x.c
[hjl@gnu-6 pages-1]$ cat x.s
    .file    "x.c"
    .text
    .align 4096
.globl foo1
    .type    foo1, @function
foo1:
.LFB0:
    .cfi_startproc
    pushq    %rbp
    .cfi_def_cfa_offset 16
    .cfi_offset 6, -16
    movq    %rsp, %rbp
    .cfi_def_cfa_register 6
    movl    $0, %eax
    leave
    .cfi_def_cfa 7, 8
    ret
    .cfi_endproc
.LFE0:
    .size    foo1, .-foo1
.globl foo2
    .type    foo2, @function
foo2:
.LFB1:
    .cfi_startproc
    pushq    %rbp
    .cfi_def_cfa_offset 16
    .cfi_offset 6, -16
    movq    %rsp, %rbp
    .cfi_def_cfa_register 6
    movl    $0, %eax
    leave
    .cfi_def_cfa 7, 8
    ret
    .cfi_endproc
.LFE1:
    .size    foo2, .-foo2

[Bug c/46859] Attribute depends on location

[davek at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46859

Dave Korn <davek at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |davek at gcc dot gnu.org

--- Comment #2 from Dave Korn <davek at gcc dot gnu.org> 2010-12-09 20:31:25 UTC ---
I note that if you re-write foo2 like this:

int 
 __attribute__((aligned(4096))) 
*
foo2a ()
{
  return 0;
}

it gets aligned properly.  I think something must be going wrong when the
parser starts to build a derived pointer type based on the underlying scalar
type, but that's just a first guess.

[Bug c/46859] Attribute depends on location

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46859

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jsm28 at gcc dot gnu.org

--- Comment #3 from H.J. Lu <hjl.tools at gmail dot com> 2010-12-09 20:45:46 UTC ---
For

int *
 __attribute__((aligned(4096))) 
foo2 ()
{
  return 0;
}

parser applies attribute to "int *" instead of foo2.

[Bug c/46859] Attribute depends on location

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46859

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> 2010-12-09 21:06:38 UTC ---
I think this is the correct behavior which is also documented this way too.

[Bug c/46859] Attribute depends on location

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46859

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> 2010-12-09 21:07:15 UTC ---
This is also why it is almost always better to put the attribute on the
declaration rather than the definition.

[Bug c/46859] Attribute depends on location

[joseph at codesourcery dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46859

--- Comment #6 from joseph at codesourcery dot com <joseph at codesourcery dot com> 2010-12-09 21:23:53 UTC ---
On Thu, 9 Dec 2010, hjl.tools at gmail dot com wrote:

> For
> 
> int *
>  __attribute__((aligned(4096))) 
> foo2 ()
> {
>   return 0;
> }
> 
> parser applies attribute to "int *" instead of foo2.

Yes, this is as documented, just like if you had a type qualifier there.  
I see that two critical words were missing in the relevant section of the 
manual; I've applied this patch to add them.

Index: doc/extend.texi
===================================================================
--- doc/extend.texi    (revision 167663)
+++ doc/extend.texi    (working copy)
@@ -4036,7 +4036,7 @@
 declaration @code{T D} specifies the type
 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
-@var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for
+@var{type-qualifier-and-attribute-specifier-list} pointer to @var{Type}'' for
 @var{ident}.

 For example,
Index: ChangeLog
===================================================================
--- ChangeLog    (revision 167663)
+++ ChangeLog    (working copy)
@@ -1,5 +1,10 @@
 2010-12-09  Joseph Myers  <joseph@codesourcery.com>

+    * doc/extend.texi (Attribute Syntax): Correct description of
+    attributes in pointer declarators.
+
+2010-12-09  Joseph Myers  <joseph@codesourcery.com>
+
     * config/mips/vxworks.h (DBX_REGISTER_NUMBER): Undefine.
     * config.gcc (mips64*-*-linux* | mipsisa64*-*-linux*,
     mips*-*-linux*, mips-wrs-vxworks): Don't use svr4.h.

[Bug c/46859] Attribute depends on location

[mpolacek at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46859

Marek Polacek <mpolacek at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
                 CC|                            |mpolacek at gcc dot gnu.org
         Resolution|---                         |FIXED

--- Comment #7 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
So nothing more to do here.