[Bug c++/18761] New: C++ ABI bug on OS X with enums

[Bug c++/18761] New: C++ ABI bug on OS X with enums

[mmitchel at gcc dot gnu dot org]

This program does not give the same alignment for A and B, which is very weird,
since the only difference is that B contains a declaration of a nested type. 
The culprit here is the Darwin round_type_align code.  I do not know if this bug
still occurs with GCC 4.0.

extern "C" int printf (const char*, ...);

union A {
  double d;
};

union B {
  enum E { e };
  double d;
};

struct AlignA {
  char c;
  A a;
};

struct AlignB {
  char c;
  B b;
};

int main () {
  printf ("Alignment of AlignA = %d\n", __alignof__ (AlignA));
  printf ("Alignment of AlignB = %d\n", __alignof__ (AlignB));
}

-- 
           Summary: C++ ABI bug on OS X with enums
           Product: gcc
           Version: 3.4.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: mmitchel at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org
GCC target triplet: powerpc-apple-darwin7.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18761

[Bug c++/18761] C++ ABI bug on OS X with doubles

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-12-01 18:56 -------
The problem has nothing to do with enums at all but doubles, if the double is the second element in 
the union it works.  I do know that that would be correct for structs but I forgot about the rules for 
unions.  Darwin uses the same rule as AIX uses for alignment so it is also a problem there too if this is 
a bug (someone will have to look it up).

In 4.0.0 we get:
Alignment of AlignA = 8
Alignment of AlignB = 4

which is what I had expected with my explaination about.

This gives the same output as the enum case (to show it has nothing to do with enums):
extern "C" int printf (const char*, ...);

union A {
  double d;
};

union B {
  int i;
  double d;
};

struct AlignA {
  char c;
  A a;
};

struct AlignB {
  char c;
  B b;
};

int main () {
  printf ("Alignment of AlignA = %d\n", __alignof__ (AlignA));
  printf ("Alignment of AlignB = %d\n", __alignof__ (AlignB));
}

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
 GCC target triplet|powerpc-apple-darwin7.4.0   |ppc-darwin, ppc-aix
           Keywords|                            |ABI, wrong-code
            Summary|C++ ABI bug on OS X with    |C++ ABI bug on OS X with
                   |enums                       |doubles


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18761

[Bug c++/18761] C++ ABI bug on OS X with doubles

[mark at codesourcery dot com]

------- Additional Comments From mark at codesourcery dot com  2004-12-01 18:59 -------
Subject: Re:  C++ ABI bug on OS X with doubles

pinskia at gcc dot gnu dot org wrote:
> ------- Additional Comments From pinskia at gcc dot gnu dot org  2004-12-01 18:56 -------
> The problem has nothing to do with enums at all but doubles, if the double is the second element in 
> the union it works.  I do know that that would be correct for structs but I forgot about the rules for 
> unions.  Darwin uses the same rule as AIX uses for alignment so it is also a problem there too if this is 
> a bug (someone will have to look it up).

Andrew, you're being careless again.  Look harder.



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18761

[Bug c++/18761] C++ ABI bug on OS X with emebed types

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-12-01 19:08 -------
oh, yes now I see it.
Confirmed, the problem is really in rs6000_special_round_type_align where we only skip VAR_DECLs 
but we should be skiping TYPE_DECLS also.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2004-12-01 19:08:28
               date|                            |
            Summary|C++ ABI bug on OS X with    |C++ ABI bug on OS X with
                   |doubles                     |emebed types


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18761

[Bug c++/18761] C++ ABI bug on OS X with emebed types

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-12-01 19:10 -------
This patch should fix it:
Index: rs6000.c
===============================================================
====
RCS file: /cvs/gcc/gcc/gcc/config/rs6000/rs6000.c,v
retrieving revision 1.757
diff -u -p -r1.757 rs6000.c
--- rs6000.c	27 Nov 2004 23:00:57 -0000	1.757
+++ rs6000.c	1 Dec 2004 19:09:39 -0000
@@ -3042,7 +3042,7 @@ rs6000_special_round_type_align (tree ty
 
   /* Skip all the static variables only if ABI is greater than
      1 or equal to 0.  */
-  while (field != NULL && TREE_CODE (field) == VAR_DECL)
+  while (field != NULL && TREE_CODE (field) != FIELD_DECL)
     field = TREE_CHAIN (field);
 
   if (field == NULL || field == type || DECL_MODE (field) != DFmode)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18761

[Bug c++/18761] C++ ABI bug on OS X with emebed types

[mark at codesourcery dot com]

------- Additional Comments From mark at codesourcery dot com  2004-12-01 19:13 -------
Subject: Re:  C++ ABI bug on OS X with emebed types

pinskia at gcc dot gnu dot org wrote:
> ------- Additional Comments From pinskia at gcc dot gnu dot org  2004-12-01 19:08 -------
> oh, yes now I see it.
> Confirmed, the problem is really in rs6000_special_round_type_align where we only skip VAR_DECLs 
> but we should be skiping TYPE_DECLS also.

Right -- and CONST_DECLs!



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18761

[Bug c++/18761] C++ ABI bug on OS X with embed types

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-11 03:51 -------
Patch here: <http://gcc.gnu.org/ml/gcc-patches/2005-01/msg00561.html>, sorry for not sending it 
out sooner, I was reminded today (at least twice) to send the patch out. 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |pinskia at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18761

[Bug target/18761] C++ ABI bug on OS X with embed types

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-01-11 20:15 -------
Subject: Bug 18761

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	pinskia@gcc.gnu.org	2005-01-11 20:15:14

Modified files:
	gcc            : ChangeLog 
	gcc/config/rs6000: rs6000.c 

Log message:
	2005-01-11  Andrew Pinski  <pinskia@physics.uc.edu>
	
	PR target/18761
	* config/rs6000/rs6000.c (rs6000_special_round_type_align):
	Skip all DECLs except for FIELD_DECLs.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.7093&r2=2.7094
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/rs6000/rs6000.c.diff?cvsroot=gcc&r1=1.773&r2=1.774



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18761

[Bug target/18761] C++ ABI bug on OS X with embed types

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-01-11 20:19 -------
Subject: Bug 18761

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	pinskia@gcc.gnu.org	2005-01-11 20:19:36

Modified files:
	gcc/testsuite  : ChangeLog 
Added files:
	gcc/testsuite/g++.dg/abi: align1.C 

Log message:
	2005-01-11  Andrew Pinski  <pinskia@physics.uc.edu>
	
	PR target/18761
	* g++.dg/abi/align1.C: New test.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&r1=1.4878&r2=1.4879
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/g++.dg/abi/align1.C.diff?cvsroot=gcc&r1=NONE&r2=1.1



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18761

[Bug target/18761] C++ ABI bug on OS X with embed types

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-11 20:20 -------
Fixed on the mainline.  Also committed a testcase so we don't fail again and find other targets which 
are wrong.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.0.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18761