[Bug c++/14808] New: Undefined results with virtual base classes

[Bug c++/14808] New: Undefined results with virtual base classes

[dannysmith at users dot sourceforge dot net]

With both 3.4.0 (20040327) and trunk (20040331), I am getting segfaults
or unexpected results with virtual base classes on i386-pc-mingw32


Here is an illustration:

// vbase.cc
#include <stdio.h>

class Base
{
public:
 // int base_var;
  virtual char* vf() const = 0;
  virtual ~Base() {}
};

class D1 :  public Base
{
public:
  char* vf() const { return "D1"; }
};


class D2 :  virtual public Base
{
public:
  char* vf() const { return "D2"; }
};


void test1()
{
  Base* b1 = new D1;
  printf("%s\n", b1->vf());
}

void test2()
{
  Base* b2 = new D2;
  printf("%s\n", b2->vf());
}

int main()
{
 test1();
 test2();
 return 0;
}

//end vbase.cc

With trunk:

When compiled without optimisation
g++ -ovbase0 -Wall -O0 vbase.cc

I get successful build, with no warnings. But executing vbase0 causes
segfault.

Ditto with -O1

When compiled with -O2 optimisation
g++ -ovbase2 -Wall -O2 vbase.cc

executing vbase2 outputs:

D1
2D2


When I add -g:
g++ -ovbase0g -g -Wall -O0 vbase.cc
Executing vbase0g doesn't segfault. But it doesn't produce _any_ 
output either.  Ditto with -O1 -g.


This is regression from 3.3.3 which output 

D1
D2

at all levels of optimisation

2.95.3 also produced the correct output.


If I add a variable (the commented out int base_var) to the
base class, the executable built with 3.4.0 and trunk executes
without problem and produces the correct output.

If I comment out the "virtual" keyword in D2's definition, I 
also get correct output with 3.4.0 and trunk.


Danny

-- 
           Summary: Undefined results with virtual base classes
           Product: gcc
           Version: 3.4.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: dannysmith at users dot sourceforge dot net
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i686-pc-mingw32
  GCC host triplet: i686-pc-mingw32
GCC target triplet: i686-pc-mingw32


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808

[Bug c++/14808] [3.4/3.5 Regression] [win32] Undefined results with virtual base classes

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
            Summary|Undefined results with      |[3.4/3.5 Regression] [win32]
                   |virtual base classes        |Undefined results with
                   |                            |virtual base classes
   Target Milestone|---                         |3.4.1


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808

[Bug c++/14808] [3.4/3.5 Regression] [win32] Undefined results with virtual base classes

[dannysmith at users dot sourceforge dot net]

------- Additional Comments From dannysmith at users dot sourceforge dot net  2004-04-05 21:59 -------
Hello

I suspect that this bug is due to the way that the PE-COFF emulation of
GNU ld handles linkonce sections.  Specifically, ld does not reliably
handle linkonce sections contining more than one global symbol. For
example, this assembler code also produces a faulting app:

	.file	"bar2.c"
	.def	___main;	.scl	2;	.type	32;	.endef
	.text
	.align 2
.globl _main
	.def	_main;	.scl	2;	.type	32;	.endef
_main:
	pushl	%ebp
	movl	%esp, %ebp
	subl	$8, %esp
	andl	$-16, %esp
	movl	$16, %eax
	call	__alloca
	call	___main
	call	__Z3onev
	call	__Z8call_onev
	leave
	.p2align 4,,4
	ret
	.section	.text$_Z8call_onev,"x"
	.linkonce discard
	.align 2
.globl __Z8call_onev
	.def	__Z8call_onev;	.scl	2;	.type	32;	.endef
__Z8call_onev:
	subl	$12, %esp
	call	__Z3onev
	addl	$12, %esp
	ret
#	.section	.text$_Z3onev,"x"
#	.linkonce discard
	.align 2
.globl __Z3onev
	.def	__Z3onev;	.scl	2;	.type	32;	.endef
__Z3onev:
	movl	$1, %eax
	ret

but is okay if I put the the the two functions in separate .linkonce sections,

Referring to the testcase in my first comments for this PR, if the
virtual function in the derived class is not inlined, and so does not
use linkonce semantics, then the app built from the modified testcase
executes correctly. Likewise, if I comment out this code in cp/method,c
(use_thunk)

*** method.c.orig	Mon Apr 05 21:49:30 2004
--- method.c	Mon Apr 05 00:41:59 2004
*************** use_thunk (tree thunk_fndecl, bool emit_
*** 409,416 ****
--- 409,418 ----
        if (DECL_SECTION_NAME (function) != NULL && DECL_ONE_ONLY (function))
  	{
  	  resolve_unique_section (thunk_fndecl, 0, flag_function_sections);
+ #if !(defined (__MINGW32__ ) || defined (__CYGWIN__))
  	  /* Output the thunk into the same section as function.  */
  	  DECL_SECTION_NAME (thunk_fndecl) = DECL_SECTION_NAME (function);
+ #endif
  	}
      }
  #endif

the app is okay.

But, as indicated by Jakub when the local aliases for virtual thunks code was
introduced

http://gcc.gnu.org/ml/gcc-patches/2003-06/msg02603.html

"If a method is emitted in a linkonce section, it will emit the thunk into
 the same linkonce section as well, since local symbols in linkonce sections
 cannot be safely accessed from other sections."

this is not really a safe thing to do.

Should _all_ the make_alias_for_thunk code in method.c be guarded by a
target macro (e.g. TARGET_USE_LOCAL_ALIAS_FOR_VTHUNK), which defaults to
1 #ifdef ASM_OUTPUT_DEF, but can be overriden by targets which define
ASM_OUTPUT_DEF, but may currently have problems with multiple globals in
linkonce sections?

Danny

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at redhat dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808

[Bug c++/14808] [3.4/3.5 Regression] [win32] Undefined results with virtual base classes

[dannysmith at users dot sourceforge dot net]

------- Additional Comments From dannysmith at users dot sourceforge dot net  2004-04-07 21:02 -------
Patch submitted:
http://gcc.gnu.org/ml/gcc-patches/2004-04/msg00432.html

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808

[Bug c++/14808] [3.4/3.5 Regression] [win32] Undefined results with virtual base classes

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-04-07 21:34 -------
Confirmed I know this is too close but still I think this should be fixed for 3.4.0.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mmitchel at gcc dot gnu dot
                   |                            |org
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
           Keywords|                            |patch
   Last reconfirmed|0000-00-00 00:00:00         |2004-04-07 21:34:53
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808

[Bug c++/14808] [3.4/3.5 Regression] [win32] Undefined results with virtual base classes

[mmitchel at gcc dot gnu dot org]

------- Additional Comments From mmitchel at gcc dot gnu dot org  2004-04-07 22:11 -------
The patch is OK for the mainline.

On the 3.4.0 branch, it is *not* OK.

Instead, please submit an uglier, but safer, patch that tests for __CYGWIN__ and
__MINGW32__ in the same place where the mainline patch checks the new target macro.

In other words:

  #if defined (ASM_OUTPUT_DEF) && !defined(__CYGWIN__) && !...
    alias version
  #else
    non-alias version
  #endif

Please attach that patch to the PR after testing.

I will then approve it for 3.4.0.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808

[Bug c++/14808] [3.4/3.5 Regression] [win32] Undefined results with virtual base classes

[mmitchel at gcc dot gnu dot org]

------- Additional Comments From mmitchel at gcc dot gnu dot org  2004-04-08 21:23 -------
This patch is OK for 3.4.0.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.4.1                       |3.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808

[Bug c++/14808] [3.4/3.5 Regression] [win32] Undefined results with virtual base classes

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-04-08 22:16 -------
Subject: Bug 14808

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	gcc-3_4-branch
Changes by:	dannysmith@gcc.gnu.org	2004-04-08 22:16:01

Modified files:
	gcc/cp         : ChangeLog method.c 

Log message:
	PR c++/14808
	* method.c (make_alias_for_thunk): Just return function decl
	for one_only functions if __CYGWIN__ or __MINGW32__
	(use_thunk): Don't put function and thunk in same one_only
	section if __CYGWIN__ or __MINGW32__.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cp/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=1.3892.2.95&r2=1.3892.2.96
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cp/method.c.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=1.275.4.2&r2=1.275.4.3



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808

[Bug c++/14808] [3.4/3.5 Regression] [win32] Undefined results with virtual base classes

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-04-08 22:31 -------
Subject: Bug 14808

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	dannysmith@gcc.gnu.org	2004-04-08 22:31:23

Modified files:
	gcc            : ChangeLog defaults.h 
	gcc/config/i386: cygming.h 
	gcc/cp         : ChangeLog method.c 
	gcc/doc        : tm.texi 

Log message:
	PR c++/14808
	* defaults.h (TARGET_USE_LOCAL_THUNK_ALIAS_P): New macro. Default
	to 1 if ASM_OUTPUT_DEF is defined.
	* doc/tm.texi (TARGET_USE_LOCAL_THUNK_ALIAS_P): Document.
	* config/i386/cygming.h (TARGET_USE_LOCAL_THUNK_ALIAS_P): Define.
	Set to non-zero iff not a one_only decl.
	
	* cp/method.c (use_thunk): Test TARGET_USE_LOCAL_THUNK_ALIAS_P rather
	than ASM_OUTPUT_DEF.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.3367&r2=2.3368
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/defaults.h.diff?cvsroot=gcc&r1=1.132&r2=1.133
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/i386/cygming.h.diff?cvsroot=gcc&r1=1.13&r2=1.14
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cp/ChangeLog.diff?cvsroot=gcc&r1=1.4030&r2=1.4031
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cp/method.c.diff?cvsroot=gcc&r1=1.279&r2=1.280
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/doc/tm.texi.diff?cvsroot=gcc&r1=1.317&r2=1.318



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808

[Bug c++/14808] [3.4/3.5 Regression] [win32] Undefined results with virtual base classes

[dannysmith at users dot sourceforge dot net]

------- Additional Comments From dannysmith at users dot sourceforge dot net  2004-04-08 22:35 -------
Fixed on 3.4.0 and trunk
Danny

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808

[Bug c++/14808] [3.4/3.5 Regression] [win32] Undefined results with virtual base classes

[dank at kegel dot com]

------- Additional Comments From dank at kegel dot com  2004-06-07 12:15 -------
Turns out this patch may have caused cygwin->linux crossbuilds to
fail, since in that case __CYGWIN__ is defined even though the
target isn't cygwin.  The symptom is
strstream.s:8390: Error: junk `(%ecx)' after expression
when building libstdc++.
See http://gcc.gnu.org/ml/gcc/2004-06/msg00429.html for a patch
that lets my builds finish.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808

[Bug c++/14808] [3.4/3.5 Regression] [win32] Undefined results with virtual base classes

[mark at codesourcery dot com]

------- Additional Comments From mark at codesourcery dot com  2004-06-07 14:29 -------
Subject: Re:  [3.4/3.5 Regression] [win32] Undefined results
 with virtual base classes

dank at kegel dot com wrote:

> ------- Additional Comments From dank at kegel dot com  2004-06-07 12:15 -------
> Turns out this patch may have caused cygwin->linux crossbuilds to
> fail, since in that case __CYGWIN__ is defined even though the
> target isn't cygwin.  The symptom is
> strstream.s:8390: Error: junk `(%ecx)' after expression
> when building libstdc++.
> See http://gcc.gnu.org/ml/gcc/2004-06/msg00429.html for a patch
> that lets my builds finish.

This patch is fine for 3.4.1; please check it in.



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808

[Bug c++/14808] [3.4/3.5 Regression] [win32] Undefined results with virtual base classes

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-06-08 06:31 -------
Subject: Bug 14808

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	gcc-3_4-branch
Changes by:	wilson@gcc.gnu.org	2004-06-08 06:30:33

Modified files:
	gcc            : ChangeLog 
	gcc/config/i386: cygwin.h cygming.h 
	gcc/cp         : ChangeLog method.c 

Log message:
	Patch from Dan Kegel.
	PR c++/14808
	* config/i386/cygwin.h (TARGET_IS_PE_COFF): New.
	* config/i386/cygming.h (TARGET_IS_PE_COFF): New.
	* method.c (make_alias_for_thunk, use_thunk): Use TARGET_IS_PE_COFF
	instead of __CYWGIN__ and __MINGW32__.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=2.2326.2.477&r2=2.2326.2.478
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/i386/cygwin.h.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=1.85&r2=1.85.10.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/i386/cygming.h.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=1.11.4.1&r2=1.11.4.2
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cp/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=1.3892.2.117&r2=1.3892.2.118
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cp/method.c.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=1.275.4.3&r2=1.275.4.4



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808

[Bug c++/14808] [3.4/3.5 Regression] [win32] Undefined results with virtual base classes

[wilson at specifixinc dot com]

------- Additional Comments From wilson at specifixinc dot com  2004-06-08 06:33 -------
Subject: Re:  [3.4/3.5 Regression] [win32] Undefined results
 with virtual base classes

mark at codesourcery dot com wrote:
> This patch is fine for 3.4.1; please check it in.

Dan Kegel isn't in the MAINTAINERS file.  I checked in the patch.  I had 
to apply it by hand.  (Dan: please use MIME attachments for patches.) 
Because the patch is small, and is essentially what I suggested, I don't 
believe any copyright assignment is needed.

In addition to fixing cygwin-x-linux cross compilers, this will also fix 
linux-x-cygwin cross compilers.  We have had a number of bugs reported 
against the latter, but I don't think anyone ever tried to track down 
the problem.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808

[Bug c++/14808] [3.4/3.5 Regression] [win32] Undefined results with virtual base classes

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-06-16 22:59 -------
Subject: Bug 14808

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	gcc-3_4-branch
Changes by:	wilson@gcc.gnu.org	2004-06-16 22:59:42

Modified files:
	gcc            : ChangeLog 
	gcc/config/i386: cygwin.h 

Log message:
	PR c++/14808
	* config/i386/cygwin.h (TARGET_IS_PE_COFF): Delete duplicate macro.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=2.2326.2.510&r2=2.2326.2.511
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/i386/cygwin.h.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=1.85.10.1&r2=1.85.10.2



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14808